And checkout the stackable alloy plates between the chips:
http://www.dddac.com/pictures/sales/sales_tool_large.jpg
That's very interesting, I hadn't known about that design going to 120 chips. With the TDA1543 needing 50mA at 5V and more at 8V that's going to call for a really serious power supply of 6A or more. Probably though it doesn't need a fan as the heat sources are all well separated. But I think putting it all in a box could be a stretch unless its a poweramp-sized one.
These past few days I've been working on power supplies and the I/V stage. Here is the prototype I/V stage (left-most board) hooked up into a basic implementation of Stack DAC :
The DAC board needs +7V and the I/V board, -12V. So, at present, two power supplies but I am looking into a way of combining those two rails into one board. In the middle is a 5th order elliptic filter board.
Since the DAC boards act as simple current sources (unipolar) the I/V stage can be as simple as just a resistor. But the resistor has to be to the -ve rail and if the output uses GND as a reference then the power supply adds some noise in bridging those two nodes. If you don't need the GND to be the output reference - for example your amp's input is trafo isolated - then you're good to use the -12V as your output 'GND' reference. But for many, GND is needed as output reference so my I/V design creates an ultra-low noise power supply locally using a string of IR LEDs and adds buffers to give a low output impedance even with a single board in the stack.
Last night we did a quick A/B between a single board and two boards (12 DACs vs 24) in the stack. Wifey said the single board sounded 'more messy' which I interpret as 'higher noise'. To my ears the bass improvement is the most noticeable, two boards gives more solidity to the ambient acoustic and seemingly more 'weight' to lower frequencies.
For those who are curious to understand what difference a filter makes, we've made a board which replaces the filter with a couple of 0R resistors so you can compare. The filter in this pic needs pre-selected inductors to get a respectable frequency response but I have a design with lower component sensitivity coming which works with Mouser inductors. Then you can build a filter yourself if you're inclined without worrying too much about inductor tolerances.
The DAC board needs +7V and the I/V board, -12V. So, at present, two power supplies but I am looking into a way of combining those two rails into one board. In the middle is a 5th order elliptic filter board.
Since the DAC boards act as simple current sources (unipolar) the I/V stage can be as simple as just a resistor. But the resistor has to be to the -ve rail and if the output uses GND as a reference then the power supply adds some noise in bridging those two nodes. If you don't need the GND to be the output reference - for example your amp's input is trafo isolated - then you're good to use the -12V as your output 'GND' reference. But for many, GND is needed as output reference so my I/V design creates an ultra-low noise power supply locally using a string of IR LEDs and adds buffers to give a low output impedance even with a single board in the stack.
Last night we did a quick A/B between a single board and two boards (12 DACs vs 24) in the stack. Wifey said the single board sounded 'more messy' which I interpret as 'higher noise'. To my ears the bass improvement is the most noticeable, two boards gives more solidity to the ambient acoustic and seemingly more 'weight' to lower frequencies.
For those who are curious to understand what difference a filter makes, we've made a board which replaces the filter with a couple of 0R resistors so you can compare. The filter in this pic needs pre-selected inductors to get a respectable frequency response but I have a design with lower component sensitivity coming which works with Mouser inductors. Then you can build a filter yourself if you're inclined without worrying too much about inductor tolerances.
Here's a weird one - I built up 6 layers of Stack DAC with the latest boards, using DAC chips that I had pre-tested. Out of the six boards, one had a strange malfunction - it would play perfectly provided I set Foobar to 0dB (max volume) but at any level below maximum it added noise. Given that Foobar does a fade in a few hundred mS when you pause or stop playback this meant each time it faded, I got a burst of noise for the duration of the fade. The even more bizarre symptom was - reducing the volume to minimum (-100dB) gave perfect noiseless sound, although rather quiet, definitely not -100dB quiet though (that would be inaudible), probably -40dB or so.
The culprit I found by shorting pin7 to GND on each DAC in turn while playing at -1dB - shorting the faulty DAC's pin7 to GND got rid of the noise whereas pin7 to GND on all others did nothing audible beyond a slight click, so I replaced that single chip and everything is now fine.
What my hypothesis is for these super-bizarro symptoms is that somehow the input logic for the chip got damaged and it was losing one or more bits of the digital input word. On my Windows when at maximum volume Foobar only puts out 16 significant bits, the remaining 8 bits are zeroes (I only play RBCD material in general). When the volume's reduced those 8 bits are no longer zero so I figure the damaged chip was sending those to the output in some way which generated noise. A working chip just ignores those bottom bits, beyond 16. At -100dB I guess only those bottom 8 bits are active (haven't checked that on a scope though) which gave quiet music from the damaged chip, all the other chips putting out silence.
The culprit I found by shorting pin7 to GND on each DAC in turn while playing at -1dB - shorting the faulty DAC's pin7 to GND got rid of the noise whereas pin7 to GND on all others did nothing audible beyond a slight click, so I replaced that single chip and everything is now fine.
What my hypothesis is for these super-bizarro symptoms is that somehow the input logic for the chip got damaged and it was losing one or more bits of the digital input word. On my Windows when at maximum volume Foobar only puts out 16 significant bits, the remaining 8 bits are zeroes (I only play RBCD material in general). When the volume's reduced those 8 bits are no longer zero so I figure the damaged chip was sending those to the output in some way which generated noise. A working chip just ignores those bottom bits, beyond 16. At -100dB I guess only those bottom 8 bits are active (haven't checked that on a scope though) which gave quiet music from the damaged chip, all the other chips putting out silence.
It was sounding very enjoyable just as a stack of 3 boards. But when I added another 3 and went over to balanced 3+3 it went up to a whole new level. Natural is how I'd describe it now, it gets out of the way of the recording more. It seems hard to believe that adding 2 more boards to each stack will bring further improvement.
Oh, I forgot to mention I upgraded the filters since I took the pic a few posts above. I now have 7th order elliptic - this slightly out-performs my older 7th order filter as seen on DecaDAC since that was quasi-elliptic and didn't manage quite as much stop-band rejection.
Oh, I forgot to mention I upgraded the filters since I took the pic a few posts above. I now have 7th order elliptic - this slightly out-performs my older 7th order filter as seen on DecaDAC since that was quasi-elliptic and didn't manage quite as much stop-band rejection.
I'm almost done populating the first PCB version of the I/V stage. It has an output level control which is a step removed from a volume control in that adjusting it while music is playing is going to induce a 'clunk' due to a change in DC offset. At least in SE mode, in balanced mode that'll be suppressed, it remains to be seen by how much.
First board is up and running, the only errors on the board I've found so far are footprints for some resistors being too large, easily bodged. I am thinking a few more TPs too, to check that the resistor arrays are all stuffed correctly. Now onto the 2nd board so I can try balanced. The level control does clunk, but not quite as badly as I imagined. In balanced mode I need to find a way to put the two level controls in parallel....
No, this board is designed for DACs outputting lots of current. A single Stack board gives 18mA p-p whereas an AD1862 gives 2mA p-p.
The one to use for an AD1862 would be my PCM56 DAC - I've been building the 2nd prototype of it this morning :
There are PCB pads for connection of an external DAC via twisted pairs. We'll do a cut-down version of this at some point, minus the DAC and interface circuitry (the left side of the PCB).
The one to use for an AD1862 would be my PCM56 DAC - I've been building the 2nd prototype of it this morning :
There are PCB pads for connection of an external DAC via twisted pairs. We'll do a cut-down version of this at some point, minus the DAC and interface circuitry (the left side of the PCB).
Nobody's yet asked about going to more than 120 chips by extending the Stack upwards. Does that mean no interest? 
The reason I haven't gone higher than 5 boards in the Stack is largely because of the filter. With more current from the DACs, the termination resistor of the LC filter needs to be of a lower and lower value to give the standard 2VRMS output. Five boards in the stack gives an I/V resistor about 64ohms. If we went to 10 boards then we'd have 32ohms for the same output level. With such a low resistance termination the inductor values become less easy to create accurately because the inductance comes as the number of turns squared and the number of turns can only be an integer. So the inductor values become more and more coarsely quantized as their value goes down. A secondary issue is the capacitor values need to go up as the Ls go down and this means paralleling more and more 100nF NP0s (220nF exists in NP0 but they're much more expensive).
Ways around this issue - choose cores with bigger airgaps. This would raise the number of turns needed for a given value of inductance. Problem is - I don't know where to find such cores and this only solves the inductor difficulty. I could go over to air-cored, this would work but aircored the inductors would be very bulky compared to the P14 cores. The easiest route at present seems to be - run at a higher output level and then have a transformer to get back to the standard 2VRMS. This way we can use the existing filter.
The output MOSFETs on the Stack boards can handle 20V max so the limit in raising the output level will be 6VRMS meaning we could get to a 15board stack if anyone was suitably inclined.
The reason I haven't gone higher than 5 boards in the Stack is largely because of the filter. With more current from the DACs, the termination resistor of the LC filter needs to be of a lower and lower value to give the standard 2VRMS output. Five boards in the stack gives an I/V resistor about 64ohms. If we went to 10 boards then we'd have 32ohms for the same output level. With such a low resistance termination the inductor values become less easy to create accurately because the inductance comes as the number of turns squared and the number of turns can only be an integer. So the inductor values become more and more coarsely quantized as their value goes down. A secondary issue is the capacitor values need to go up as the Ls go down and this means paralleling more and more 100nF NP0s (220nF exists in NP0 but they're much more expensive).
Ways around this issue - choose cores with bigger airgaps. This would raise the number of turns needed for a given value of inductance. Problem is - I don't know where to find such cores and this only solves the inductor difficulty. I could go over to air-cored, this would work but aircored the inductors would be very bulky compared to the P14 cores. The easiest route at present seems to be - run at a higher output level and then have a transformer to get back to the standard 2VRMS. This way we can use the existing filter.
The output MOSFETs on the Stack boards can handle 20V max so the limit in raising the output level will be 6VRMS meaning we could get to a 15board stack if anyone was suitably inclined.
I suspect the power supply and cost would be a big factor in going 120+ chips. Those points aside I would love to find out what is on the other side!
After reading the DDDAC website, the thought of huge amount of chips became very interesting. They mention on the site that adding more chips creates more "bits" and by adding 256 chips one could gain 4 bits than usual. Seems a long road for 4 bits but nevertheless fascinating.
A good question would be given your experience with the stacked boards Richard, would you try 120+? Do you think there is more top be gained, from a sound perspective?
After reading the DDDAC website, the thought of huge amount of chips became very interesting. They mention on the site that adding more chips creates more "bits" and by adding 256 chips one could gain 4 bits than usual. Seems a long road for 4 bits but nevertheless fascinating.
A good question would be given your experience with the stacked boards Richard, would you try 120+? Do you think there is more top be gained, from a sound perspective?
Wifey's been working on the layout for the PSU which can cope with 120 chips (1.2A) - its using a switcher followed by an LC filter and a linear regulator to reduce the output noise. Without the switcher, the heatsink for the regulator would have to be much bigger. You'll need to supply a 20VA mains trafo to feed it with AC. Of course that's only one of the two PSUs required, the -12V one though doesn't have such a heavy current requirement.
As regards adding bits, yes and no. The audio being fed to the DAC is the determinant of the number of bits in one sense, but as regards the ultimate noisefloor, that does go down by the equivalent of one bit for every quadrupling of the number of chips. So indeed 256 chips gives a 4bit lower noisefloor. As to whether I'll try, probably yes as I have some of the earlier generation boards lying around, might as well put them to good use eh? Hard to say how much is still being left on the table at the 120chip level until we've experienced substantially more
As regards adding bits, yes and no. The audio being fed to the DAC is the determinant of the number of bits in one sense, but as regards the ultimate noisefloor, that does go down by the equivalent of one bit for every quadrupling of the number of chips. So indeed 256 chips gives a 4bit lower noisefloor. As to whether I'll try, probably yes as I have some of the earlier generation boards lying around, might as well put them to good use eh? Hard to say how much is still being left on the table at the 120chip level until we've experienced substantially more
Wifey's been working on the layout for the PSU....
The boards have come back and I've built up a couple - here's one of them :
There's a new kind of common-mode choke (that green bar in the middle is its core) which uses flat wire. Definitely a step forward in measured performance compared to the older type with traditional round wire. The three '100' inductors are needed to spread out the heat from the low pass filter - originally I had just a single 33uH inductor but it was getting so hot I figured it should be split into 3. A certain DCR is required as damping for the filter, changing to a lower ESR inductor would introduce unwanted peaking.
Here's the first prototype stand-alone I/V stage. What this is, is a PCM56 DAC minus the PCM56s (and digital interface). It has a +/-1mA current input to a cascoded common-base (discrete) stage, followed by a passive filter (not fitted for the pic). You can use a wide selection of R2R DACs (AD1862, PCM1702/4 and the like) and have strong attenuation of the images in NOS from the 7th or 9th order LC filter which sits on top. Whilst the filters for this are mechanically compatible with DecaDAC and Celibidache, they're electrically different due to the I/V resistor on this board being a much higher value (1500ohm). For those who love MELFs, there is provision on this PCB to use 1k5 MELFs for I/V duty.
Eagle-eyed viewers will note there are unfilled LED footprints - that's because I plan to try running this at higher supply rails to get even lower noise from the current sources, +/-18V rails is how its set up at present.
Last night I rigged up a couple of TDA1387s on a perf-board to test this out and its the best sound I've ever had from just 2*TDA1387s. Of course its noisier than my multiple-paralleled DACs but putting that aside, it still sounds highly enjoyable. To run this board with an array of TDA1387s would require a step-up trafo to reduce the current swing. Hmmm.....
