Open Source DSP XOs

[twest820]

Originally Posted by mhelin

LM1117-5.0/3.3 LDO's a fine for regulating the DAC voltages, aren't there any reasons for not using them?

The 1117 family is a good default for a positive LDO and ON's NCP1117 parts are probably the most useful implementation thereof; cheap, readily available, available in TO-252 if you need the thermals, and ceramic stable with sufficient output capacitance. Asus specs the H6 as having three PCB1796 DACs, which I assume is a typo for PCM1796. TI doesn't specify the PCM1796's PSRR or provide eval boards for the Burr-Brown DAC family so it's hard to estimate whether the chip's headline DNR spec is supply limited---given a typical DAC PSRR as a function of frequency -120ish dBA is about where things wind up with any of the many Vrms_noise = 0.003% Vout regulators when used with minimal supply filtering. Also, it's common for manufacturers to market their boards as having the DAC's datasheet performance. So, unless you can find actual measurements, it's reasonable to expect the H6 may deliver performance anywhere from a few to many dB worse than the PCM1796's 123dBA DNR datasheet spec. Also, note that a DAC's headline figures are typ and min values aren't usually specified. So if you get unlucky with the DAC you end up with performance may be several dB worse as well.

So, while the TI parts qusp mentioned are nice, they may or may not result in wringing a better THD+N or DNR out of the H6 than a 1117. It's most likely performance will be limited by Asus's PCB but there's no harm in throwing extra regulator at it if you don't mind the additional cost and soldering complexity (though, depending on the gain structure and noise reduction downstream of the DAC board, there may not be any point---it's worth making sure that's capable of handling sub-uV levels before bothering with fancy regulators). Most low noise regulators don't support 5V outputs and therefore don't work with most DACs so TI's updates of the 317, 337, 317L, and 337L end up being the go to choice if one wants something better than a 1117. The ADP7102/7104 are also potential choices for this application and TI has a few other parts which might be of interest---you can go to TI's part selector and filter for low noise but probably the most interesting one in this case is the TPS71701.

[qusp]

sure I agree you have to make sure the board is up for it and there are lots of choices (the ubiquitous lt1763A is not a bad one either), but if building an external reg board as a sort of test fixture and with the availability of soldering technique, a suitable PCB and equipment, the cost is still pretty low to build something that will warranty that if there are noise issues, its with the dac PCB.

thats just me though and the same parts could be used for an analogue supply upgrade for the dac directly. a DIY 2 layer PCB is still a possibility at these sizes if you are organized and prototype quantities from board houses are so cheap now.

its just nice to know now that unless chasing really high specs, the ability to use a pretty simple LDO based supply for both digital and analogue is doable, vs having to do something like a flea type buffered reference (for the space and complexity constrained)

the flexibility of the 1A models is excellent, enabling you to power anything from a clock or 3v3 logic, right up to a pretty serious class A discrete fet or bipolar based trans-impedance amp at +/-20-30v.

[twest820]

Well, they can be flexible, but the lack of TO-263 or TO-220 availability imposes thermal limits on what one can do with them---~40 C/W is kind of a painful package for a 1A regulator. Ripple rejection is another significant concern; for example, if one really wants to realize the 7A4700's 4.7uV RMS noise floor the audio band ripple on the regulator's input needs to be under 5mV or so. As a result, a preregulator is pretty much required, both to manage the ripple and provide sufficient line regulation. Similarly, all loads the regulator supplies have to be class A or really close to it, otherwise the rail stability limits on load regulation.

So it's been my finding it's often more effective to make the general purpose supply a 317+337 pair (or similar) in TO-220 or 263 with the option of adding heat sinks and then use the fancy parts for point of load regulation. Point of load tends to make ground loops more manageable too, which can be useful when one's trying to minimize crosstalk between digital +2.5/3.3/5V supplies, a DAC's +3.3 or 5V analog supply, and +-5V or whatever for the output buffers.

[qusp]

yeah well I always use preregulation anyway, anything shooting that low with noise should really be using it, or at least a good stiff filter. the package is both blessing and curse I agree, but i'll give Ti the benefit that its doable. I get the feeling that with some of the crazy parts packaging i'm seeing coming online lately, PCB heatsinking is more effective than you/we think, if the design is good.

I guess i'm assuming as much overkill as I have a tendency for and decent onboard regulation and decoupling. transient response actually seems pretty good so far, not sure about the neg part yet, only just got samples and working on a PCB now. to lower my own anxiety about the heatsinking i'm allowing for a BGA heatsink to be mounted to copper pour underneath with plenty of vias and then flipping the whole board upside-down

[twest820]

The ~40 C/W number I mentioned is from figure 28 in the 7A4700 datasheet and a generous allowance of PCB area (most of my boards would be in the >60 C/W range). Prolly the highest supply one would use the 7A4700 or 7A3301 would be a 20V trafo, resulting in a typical scenario of dropping around 25V to 15V if the mains are at 100%. The maximum current the regulator can handle in this configuration is something like (150C max junction temperature - 40C ambient temperature inside case) / ((25V unregulated input - 15V regulated output) * 40 C/W theta_ja) ~= 275mA, reducing to 170mA at 110% mains. If one targets a more realistic junction temperature of 100C to allow some margin and improve longevity those numbers reduce to 150 and 94mA. Kind of makes one wonder why the 4700 isn't available in TO-220 like the 3301 or why exactly one's paying for a 1A capacity.

Preregulating to, say, a 1V dropout pretty well solves the thermals, though, and seems to be how TI intends the parts to be used. This makes it a little odd both the positive and negative parts support a 36V input---for a positive regulator this makes sense in a single, preregulated supply application but for negative the only scenario I see for 36V is direct operation from a ~20V trafo. That might explain the 3301's TO-220 availablity but doesn't explain the 4700 being QFN only. Perhaps TI had customers that didn't know how to implement an inverting buck regulator or a dual positive configuration?

Be interesting to see what theta_ja you get with the BGA sinks.

[qusp]

i'm not replacing PCB heatsinking with the BGA, just augmenting it to make myself feel better. to be clear i'm talking about attaching it to the copper with something like sekisui thermal tape, not the part. yeah the to220-7 only in the neg version is a bit annoying, but i'm used to seeing this sort of crap, i'm just happy to see a pair in the same package at all, even a pair at all.

i'm planning to run them higher voltage than 36, about +/-45vdc @ 250-300ma by elevating the ground by 20-25v with a diode and returning everything relating to the reg, voltage dividers etc to that instead of ground, this way emulating 20-25v operation. all very experimental, it may end in a puff of blue smoke.

I did think of doing dual positive, but I prefer not. it is supplying a Class A IV stage, so regulation isnt such a big deal, just lower noise without breaking out the massive heatsinks to power an IV that already runs 50W

[mhelin]

Good discussion, guess diy folks when compared to professionals totally almost forgot the PSU issues and don't spend enough time for studying thing like regulators instead of looking DAC or ADC part with the lowest SNR.

Now, for prototyping work, or for using in actual 'product' it would be nice to have a general purpose power supply board with dual analog supplies for op amps, and single supplies for various bits on converter board (DAC digital and analog supplies, digital buffers and everything else). The QNF parts are difficult to solder, at least the ones with the exposed thermal pad, so other alternatives are welcome. Noticed though that TI's got EVM's for TPS7A4700 (and the 7A33 neg part), but those are about 20 EUR/pc. There is also the 150mA family of ultralow-noise regulators: TPS7A49/30:
Ultra low noise low dropout ( LDO ) voltage regulator IC for power precision analog and noise sensitive applications - TPS7A30 - TI.com

For DAC like PCM1691 whose max. Icc and Idd are <110 mA these would be ok if you had an extra one for other chips onboard. For PCM1796 the Icc and Idd are less that 20 mA, so these might work for the Asus board as well (but there are other chips using the same supplies so propably not). Well, using regular easily available 317/337 parts is one possibility.

Here in diyaudio there are also other threads regarding regulators:
Improving the LM3x7 regulator circuit

Hifiduino pages also list several LDOs:
New Breed of Ultra Low Noise Regulators H i F i D U I N O

Anyway if someone could desing a board with enough TPS7A4700 chips plus single TPS7A3301 and could order the board with the QNF parts soldered and organize a group buy for it I would order couple of boards. Then why not include also rectifiers and other components (mostly capacitors) on the same board. On some group buy (I2S FIFO I think) there is already some TPS7A4700 board, but it contains just single chip.

[TheShaman]

Small size, low noise bipolar supplies?! Sign me up. I've been looking for something like that for ages.

[qusp]

erm.... cough =) as I said i'm in the middle of designing that....

mhelin: the lower current ones are the ones I linked as well.

but no I wont be having the rectifiers on the same board, I would prefer to leave the DC input flexible so it can be batteries for low voltages. they need a fair bit of smoothing and for that and at higher voltages, that means through-hole, or it gets rather expensive or downright undoable.

I want to leave the option for just sticking it to the floor with some keratherm or sekisui and even short snipped off leads will foul that connection. so i'll probably do a matching bridge, or people can preregulate it etc. as far as ready-made modules though... that wont be the first thing I do, I need to test these things out, ive already got some of Ians little boards, but they are pretty limited as far as heatsinking because they are so tiny, they werent designed for more than 100-150ma

[qusp]

smallest size though, that would be the LT3032, the ones I used on my portable ESS dac-amp that you might have seen, all in one tiny DFN14 chip.