You only need one winding. The reason for two inductors is because the PCB gives two options - either ready-made Mouser-sourced inductors or DIY ones using the P14 cores.
I believe so, I will check back against my old schematics.
No, half-gross means only 72 chips and hence needs quite a few more turns on the primary. If my memory is correct it was around 15+15t, I need to look back.
That's right so as to give the option of SE or balanced output.
Good luck with winding and do take care not to give yourself RSI like I did!
From a quick fiddle in LTspice, the P14 inductors are 12mH for a half-gross version and the I/V resistors 910ohm. With AL160 cores the right amount of damping is given by using 0.07mm dia wire (274 turns) which gives an ESR in the region of 28ohm.
I've consulted my notebook from 2020 and found that for the Q-gross I went with 5400 turns of 0.04mm on the secondary, 250 turns of 0.16mm on the primary. So to build a H-gross you'll increase the turns ratio. This means in practice 126 turns on the primary. All windings CT'd.
If you're going for 4000 turns on the secondary then the primary will reduce in proportion, to 94 turns.
Thanks for the information!
Did you see how many turns you had for the outputs by any chance?
I'll try not to give myself RSI... 🙂
Did you see how many turns you had for the outputs by any chance?
I'll try not to give myself RSI... 🙂
My note keeping was too messy to find that so I went back to first principles.
Totally 800 turns of 0.17mm. Wind 200T for one half of the secondary. Then 400T for primary. Finally 200T for second half of the secondary.
Totally 800 turns of 0.17mm. Wind 200T for one half of the secondary. Then 400T for primary. Finally 200T for second half of the secondary.
2nd generation GrossDAC
The last few days I've been working on improving the original GrossDAC design - there's a new DAC board where individual shunt regs are used per pair of DAC chips. And on the I/V filter board I'm experimenting with an off-the-shelf transformer.
In this first prototype the output transformer isn't needed as its feeding my monoAMPs which come already equipped with input trafos. The filter boards are 5th order but I'm using up some prototype 9th order boards which for some obscure reason I have too many of.
The eagle-eyed may notice a small green prototype board in the middle - this comes with two IR LEDs to provide a mid-point bias for the centre tapped trafos. I'm running the DAC offset current through the transformer at the moment which may turn out not to be optimum.
The last few days I've been working on improving the original GrossDAC design - there's a new DAC board where individual shunt regs are used per pair of DAC chips. And on the I/V filter board I'm experimenting with an off-the-shelf transformer.
In this first prototype the output transformer isn't needed as its feeding my monoAMPs which come already equipped with input trafos. The filter boards are 5th order but I'm using up some prototype 9th order boards which for some obscure reason I have too many of.
The eagle-eyed may notice a small green prototype board in the middle - this comes with two IR LEDs to provide a mid-point bias for the centre tapped trafos. I'm running the DAC offset current through the transformer at the moment which may turn out not to be optimum.
Attachments
Last edited:
The 2nd generation prototype sounded acceptably good but thinking a bit more about lowering the noise I wondered if using a cascode MOSFET rather than the input trafos wouldn't be a lot simpler. I'd been playing with a MOSFET in a similar position in my experiments with enhancing the original PhiDAC design. In that instance it was primarily to provide level shifting, but on GrossDAC it can provide impedance transformation with level shifting as a bonus.
I am keeping the 1k resistors for termination of the filter as I'm way too lazy to redesign the filter and wind another set of inductors for it. This though means with the raw output current from the DAC board of 18mA p-p that the swing into the resistor becomes rather large - 18V p-p. But its acceptable as I have an output transformer to bring the voltage down again. In fact I just have to increase the step-down ratio into my amps from ~6 to ~36. The LM317 needs re-configuring to give a 24V supply and the reference (for mid-rail) goes up to 13V. The buffers needed their bias reducing so as not to overheat.
(more to come)
I am keeping the 1k resistors for termination of the filter as I'm way too lazy to redesign the filter and wind another set of inductors for it. This though means with the raw output current from the DAC board of 18mA p-p that the swing into the resistor becomes rather large - 18V p-p. But its acceptable as I have an output transformer to bring the voltage down again. In fact I just have to increase the step-down ratio into my amps from ~6 to ~36. The LM317 needs re-configuring to give a 24V supply and the reference (for mid-rail) goes up to 13V. The buffers needed their bias reducing so as not to overheat.
(more to come)
Here's the 2nd prototype DAC board for the design that has the place holder name 'DenaDAC'. The hand-wired interface PCB at the back holds 8 current sources (4 up, 4 down) and 4 MOSFETs to cascode the DAC outputs and give the potential for a big voltage swing.
I found an old pair of EE35 transformers which I'd originally wound as input trafos for GrossDAC, they're now being used as output transformers with 40:1 step down ratio. Which turned out to give the right drive to a TDA8932.
The resulting SQ is the best I've ever heard from TDA1387, completely addictive.
I found an old pair of EE35 transformers which I'd originally wound as input trafos for GrossDAC, they're now being used as output transformers with 40:1 step down ratio. Which turned out to give the right drive to a TDA8932.
The resulting SQ is the best I've ever heard from TDA1387, completely addictive.
Attachments
PCB revision
The updated 36 DAC board (with cascode FETs included) is being built. This one uses multiple paralleled FETs to spread the load (since the FETs are those tiny SOT-23s) then in theory I should be able to go above the 18V output swing I achieved on the earlier prototype. Whether that'll result in a gain in SQ is what this experiment is all about 🙂
Can't usefully go any higher than 36 chips on a single board as we've reached the maximum cheap board size with this one (10*10cm). Still the capability exists to wire a couple of these in parallel.... I'm not at present planning going beyond 36V swing as I don't have thin enough wire for the primary of the trafo for that so an extra board allows the I/V resistor to scale down in value.
The updated 36 DAC board (with cascode FETs included) is being built. This one uses multiple paralleled FETs to spread the load (since the FETs are those tiny SOT-23s) then in theory I should be able to go above the 18V output swing I achieved on the earlier prototype. Whether that'll result in a gain in SQ is what this experiment is all about 🙂
Can't usefully go any higher than 36 chips on a single board as we've reached the maximum cheap board size with this one (10*10cm). Still the capability exists to wire a couple of these in parallel.... I'm not at present planning going beyond 36V swing as I don't have thin enough wire for the primary of the trafo for that so an extra board allows the I/V resistor to scale down in value.
Attachments
Its been rather a while since I last posted but progress in getting the 'Stack DAC' playing engrossing music has encouraged me along the path of exploring more paralleled chips and led me back to giving this board another shot. I had shelved it due to MOSFETs continually oscillating and/or blowing up but now I figure I probably just had a dodgy reel of FETs. So I'm replacing the 2N7002s with Toshiba SSM3K335s and going to give it another try. Plus I plan to 'overclock' the DAC chips with the pull-up resistor to pin7 to eke out a little more juice...
I apologize beforehand if I seem impertinent in posing this question. I examined the provided schematic for the DenaDac TDA1387 boards, and I noticed that you do not have any passive part connected to pin 7 of the TDA1387. I am accustomed to seeing other designs with a capacitor from pin 7 to ground. Even the datasheet provides for a 1uF cap connected to pin 7 to ground. Will your scheme work without a cap to pin 7 ? Has the DenaDac v2 with the gerbers provided functioning? If it is functional, would you care to explain why pin 7 has been left unconnected?
My DF1704 oversampling experiment with the similar 1545 is interesting. The df1704 will output 16 and up to 24 bit WL's .
Does the 1387 begin to sound "degraded" above its default 16 bit WL input?
Does the 1387 begin to sound "degraded" above its default 16 bit WL input?
Not in my experience, no. I've tried it with both 24bits and 16bits from Windows and - no difference.
Looking through some of my many boxes of boards from old projects, I discovered I had rather a lot of the original 36-chip DAC boards stuffed full of TDA1387s. I could feel they all wanted to be playing music music together rather than gathering dust so I decided to revive the project and this time aim a bit higher. The target is 1000 DAC chips in a single DAC so I've decided to call it 'kiloDAC' (or kDAC for short).
Since there are 36 TDA1387s on a single board this target calls for a stack of boards 28 high, but I'm going to make it a balanced DAC so there will be two mono towers each of 14 boards. The power supply requirements are helped considerably by building a dual-mono DAC - since each chip draws around 5mA, a 500-chip stack will pull 2.5A which is within the remit of a 3A LDO like LM1085. It'll need a fairly substantial heatsink though, I plan to feed the LDOs with buck regs to minimize their heat generation. Since in the last iteration the trafos generated too much LF distortion, they're out now, just simple passive I/V will be the order of the day, preceded by a 7th order filter. I plan to include a software 2X upsampler which makes the reconstruction filter design that much less onerous.
With so much current (0.5A peak) going into the I/V resistors they'll need to be specially selected ones, I'll do a search on Taobao to see if there are any fancy ones in low enough values. I figure the I/V resistors will need to be in the region of 6ohms so we might need several in parallel.
Here's one of the 14-board stacks undergoing evaluation. If I decide to run them at their max 5.5V supply probably I'll need a slowish fan to circulate the air.
Since there are 36 TDA1387s on a single board this target calls for a stack of boards 28 high, but I'm going to make it a balanced DAC so there will be two mono towers each of 14 boards. The power supply requirements are helped considerably by building a dual-mono DAC - since each chip draws around 5mA, a 500-chip stack will pull 2.5A which is within the remit of a 3A LDO like LM1085. It'll need a fairly substantial heatsink though, I plan to feed the LDOs with buck regs to minimize their heat generation. Since in the last iteration the trafos generated too much LF distortion, they're out now, just simple passive I/V will be the order of the day, preceded by a 7th order filter. I plan to include a software 2X upsampler which makes the reconstruction filter design that much less onerous.
With so much current (0.5A peak) going into the I/V resistors they'll need to be specially selected ones, I'll do a search on Taobao to see if there are any fancy ones in low enough values. I figure the I/V resistors will need to be in the region of 6ohms so we might need several in parallel.
Here's one of the 14-board stacks undergoing evaluation. If I decide to run them at their max 5.5V supply probably I'll need a slowish fan to circulate the air.
Here's my first attempt at a filter - the I/V resistor is 5.7ohm to fit the need for an integer number of turns on the inductors. They're being wound on P18 ferrite cores. Note since the filter's terminated on both input and output the effective I/V resistance is half of 5.7ohm. I originally thought I'd need to go to air-cored inductors due to the high current involved but air-cored inductors need to be huge to get high enough Q - low Q inductors give drooping frequency response at higher frequencies. So ferrites it is for now, with dual windings to make terminating the wire ends a bit easier. Cap values are rather impractical for them to be made up out of NP0 100nFs which have been my staple for filters up until now. Larger NP0s have become available (220nF and even 470nF) but they work out much more expensive per uF. So on order are some MKPs (film caps) from Taobao.
!!!!!!!!!!!!!!!!!!!!!!!
Well, back to Earth, with so many chips aren't your afraid of delays due to chips being so far apart?
Patience must be your second name, 1000 DAC chips, bloody hell!
Well, back to Earth, with so many chips aren't your afraid of delays due to chips being so far apart?
Patience must be your second name, 1000 DAC chips, bloody hell!
Good to see you again Claude! I was half-expecting you to ask why, after such a long silence in respect of TDA1387, I'd returned to it once more.😎
Delays of propagation are going to be in the realm of nS so quite small in the scheme of things where one sample is 22,000nS. I'm packing the chips as close together as I can while allowing a little bit of air to flow having been bitten by the heating issue in the 'Stack DAC' project a couple of years back. The chips I'd mainly soldered in years gone past, I only had to solder up 4 new boards to reach 1008 chips.
Delays of propagation are going to be in the realm of nS so quite small in the scheme of things where one sample is 22,000nS. I'm packing the chips as close together as I can while allowing a little bit of air to flow having been bitten by the heating issue in the 'Stack DAC' project a couple of years back. The chips I'd mainly soldered in years gone past, I only had to solder up 4 new boards to reach 1008 chips.