16bit octal A_out

[jojjelito]

I was investigating DACs due to circumstances involving analog synthesis and came across the pin-compatible DAC8568 (octal 16-bit DAC, 10uS settling time at worst) and the AD5668 with similar characteristics. It looks like any of these two could be mounted on a PCB in order to attain 16-bit resolution with hopefully better precision than the current Aout_NG (which isn't that bad!). The datasheets don't guarantee better than 12-bits resolution at worst case but at 14.81€ (Digikey) for the AD, and at 22€ (Mouser) for the TI DAC it's not worlds apart from the TLV5630 at 18.13€ (Mouser).

I was thinking of getting a couple of each in order to do some experiments. Any thoughts about the worse settling time, or if this endeavor would be at all worthwhile? I was thinking it could be fun to write a SPI driver for these 2 ICs in the long, dark of fall.

Cheers!

[Davo]

Are you saying either of these chips can substitute for the DAC in the AOUT_NG? What makes "worst case" such that only 12 bits are honored? This sounds like a worthy line of investigation. I have one of the first generation of AOUT boards that'll eventually be used to interface to some analogue stuff. Maybe it's time for me to get an AOUT_NG board.

Edited August 22, 2010 by Davo

[jojjelito]

Are you saying either of these chips can substitute for the DAC in the AOUT_NG? What makes "worst case" such that only 12 bits are honored? This sounds like a worthy line of investigation. I have one of the first generation of AOUT boards that'll eventually be used to interface to some analogue stuff. Maybe it's time for me to get an AOUT_NG board.

Haha, unfortunately they're not pin-compatible with the TLV5630 used in the AOUT_NG :tongue: That would have been too convenient for someone else to not have noticed already. There are two families from TI and AD who just happen to share the same pinning, characteristics etc which the two chips are part of. I was planning to investigate an AOUT_16 if there is some additional performance/precision to be had, but it will require new boards with a redesigned footer for these chips. I'll whip something up for kitchen etching for myself at first :ninja: Then I'll have some fun with a couple of VCOs in the form of a couple of different CEM chips and a discrete transistor thingy from the bakelite, metal and wooden end cheeks-era.

When I was quickly skimming the datasheets I saw that the relative accuracy is 4 bits LSB of the TI DAC. The AD DAC also comes with some inherent INL error, but you're bound to see those on any DAC.

[seppoman]

I'm familiar with the TI DAC and planning to give it a try on a non-Midibox project one day as the datasheet looks quite promising. Didn't know the AD part so far. Comparing DAC specs is a very difficult task as there's always differences in measuring methods between the different companies, some specs are missing from some datasheets or given in different units or entirely different test conditions etc. From what I read, I'd say the TI part is better, especially looking at the AC characteristics like digital and analog feedthrough etc.

settling time - that's irrelevant for CV generation. It would only start to become a problem if you'd want to generate a signal with several hundred kHz bandwith.

For use in a NG successor/improvement, both parts have two major drawbacks. The first and most important one (at least if you'd want to not only use the module yourself but also give it to other people): TSSOP package! That's the same pin size and spacing like on the STM32. While I don't have any problem with soldering these, this package is definitely not at all newbie-compatible.

The other issue is that both require 32 bit writes instead of 16 bits on the other AOUTs. This is probably OK on the Core32, but on a MIOS8 PIC with its bit-banged SPI output and performance already driven to the limits, it means double the effort on each channel update.

Then you'd first need to find out (or better ask TK) if the internally processed data on the usual apps would even provide 16 bit data or not. Without the required data depth, using a 16 bit DAC makes no sense. But I think having 16 bits DAC would not give much benefit anyway. From the datasheet I'd say a real precision of the DAC output isn't much better than real 12 bits anyway, not only worst case but also in normal life, and in a usual DIYer setup the whole ecosystem like clean power, good layout (i.e. no single sided homemade PCBs), VERY short distance connections, shielded wiring, ground distribution etc is usually just not good enough to backup this precision, i.e. to have a general noise level in the system that isn't louder than any small value change. I guess maybe using the 12 bit model to save some money and avoid the need to care for different data generation/processing would make more sense. This way you'd get a more precise 12 bit signal from the better parts, of course without the coolness factor of being able to claim 16 bits precision.

S

[jojjelito]

Mmm, figures... Since I was already under the impression that the performance benefit was marginal at best I guess this one was killed in the bud :whistle: With some luck we'll find better alternatives for higher precision later on. The deal with 16-bit for me is in VCO control and in direct wavetable synthesis if one is so inclined. The added precision does nothing for VCF/VCA etc control. Maybe there's some I2S-alternative out there?

As for the homemade board this was intended just for experimentation and hardware-bringup. Once the driver is done and preliminary performance data obtained with my trusty Tektronix then it would be time for a proper board. Noise issues and layout tricks will be far more critical with these multibit monsters.

Summing it up: 12-bits it is for now. TSSOP vs SOIC is not going to go away, unfortunately. If you can mod a PS2 then you'll be fine with TSSOP. I'll be surprised to see anything but bump-grid (BGA) nastyness a few years down the road for mass-market applications. Then we DIY peeps will be SOL. I guess more cards will go with pre-soldered bits and pieces.

- PIC core, is anyone using those for new MIDIbox projects? Not to knock them, but where's the USB firmware download on those old things :frantics: It's been a good ride with MIOS8, but times are changing.

Thanks Seppo!

/J

[nILS]

Maybe there's some I2S-alternative out there?

So far TDA1543, PCM1725 and PCM1754 have been successfully used :) Also, the stm32 has a 2-channel 12bit DAC on board that works pretty well (I am using that as an audio out for the sid emulation).

PIC core, is anyone using those for new MIDIbox projects?

Yes, me with the kaffeSEQ :)

[jojjelito]

So far TDA1543, PCM1725 and PCM1754 have been successfully used :) Also, the stm32 has a 2-channel 12bit DAC on board that works pretty well (I am using that as an audio out for the sid emulation).

Yes, me with the kaffeSEQ :)

I guess we're pretty well covered in the stereo 16-bit area :sorcerer: Then there's only the mythical quest of finding a decently accurate multi-channel solution for VCOs without resorting to PCM1680 ΔΣ-type devices. Or be happy with 12bits since it's analog with some inherent instability and therefore we don't worry too much about precision...

Gotta read up more on the I2S synth instead of just browsing the repository one of these days. Plus check what the kaffeSEQ is doing. Sounds like replacing filter, grinding beans, heating water, pumping and... STOP daydreaming!