TDA1387 x8 DAC: let's check its design, mod it -or not-, play music -or not! :(-

The only DAC chips I know with diff outs are S-D ones, kinda OT for this thread which is about getting the best out of the TDA1387.

Here's a filter I'm going to try after the opamp - its designed to do two things. First correct for the NOS droop and second attenuate OOB images produced by the DAC. The filter that's currently on the opamp output (220R, 220pF) has a -3dB over 3MHz so isn't doing much (if anything at all) to tame the DAC's images. My filter's 2 pole and it provides 24dB attenuation at 100kHz. The inductor's a Coilcraft MSS1210. The 50k resistor isn't part of the filter itself, its my volume pot :)
 

Attachments

  • LeeNOSfilter.jpg
    LeeNOSfilter.jpg
    17.5 KB · Views: 1,187
  • LeesNOSplot.jpg
    LeesNOSplot.jpg
    81.7 KB · Views: 1,167
Last edited:
I've been listening now for an hour or so and this filter's definitely for keeps. The rise in HF response makes it sound just a tad more transparent and I'm getting more of an 'in focus' sound, closer to my reference DAC. Here's a hasty sketch of how to connect the filter (pink lines although only one RC network is shown in the sketch). I took out the square blue coupling caps as I already have input caps on my active speakers. If you need to preserve AC coupling move these caps to the output of the LC filter board - the Coilcraft inductors sit (face down) on my board where these caps used to be.

The resistors ringed in yellow are those to reduce to get a reduced supply voltage to the opamps. The green marks indicate the resistors feeding the shunts which are to be increased - add a 1W 100R in series with those already there by standing the existing Rs up vertically and placing the new Rs alongside.
 

Attachments

  • LeeNOSOPfilter.jpg
    LeeNOSOPfilter.jpg
    154.2 KB · Views: 1,161
Last edited:
Member
Joined 2008
Paid Member
OK, you guys (well, mostly abraxalito) have convinced me I should try one of these. :) The only sellers I can find are based in China (Taobao/Cart100), which is basically shut down until Feb 25. Anyone know of any non-Chinese sellers?

I think I follow the mods being discussed here (and also on abraxalito's blog), but I might need some hand-holding. Hopefully you're all still interested in the board when I finally obtain one!

Also, is the PCB of decent-enough quality to withstand mediocre soldering skills?
 
Hi Matt,
I'll still be here, right now I've done nothing ... but surviving a flu.
You'll try, that's nice as the meaning of the word still means the possibility of discard the item, in case it's not up to your expectations. I can't compare it to anything now so won't tell you it's fantastic or compare to billion dollar stuff or things like that.
That being said I'm a little bit worried about soldering stress of the PCB, Richard when you removed through-hole parts did you notice any weakness like vias going loose or tracks broken?
 
Member
Joined 2008
Paid Member
Malefoda: sorry to hear you have the flu, hope you get over it quickly! Knock on wood, I've escaped the flu for many years now...

A random question: from looking only at the pics, combined with my just-enough-to-be-dangerous level of knowledge, it seems the PCB is fairly simple, especially if you ignore the power supply and USB/spdif input sub-circuits. Assuming I'm not too far off the mark, how hard would it be to design a "proper" PCB for the tda1387 DAC ICs (like they are doing with the tpa311x chip over in the class d amp forum)? Something like the Subbu DAC, a modular component that you can build as simple or complex system around?
 
Lower C42 sounds good to flatten the peak. Any counter-effect?

Why do you want to flatten the peak? Are you running at 192kHz?

The peak is there intentionally because when running at 44k1 the DAC chips don't give a flat frequency response, they 'droop' about 3dB by 20kHz. There's an old post on my blog about this. So this filter is specifically designed to undo that drooping effect and give a flat response overall. If you really want a flat filter then I'll design one for you - just increasing C42 will certainly flatten it but might lower the corner frequency too, so you lose the highest freqs.

I omitted to mention that this filter's only suitable for use with RBCD source material - i.e. stuff sampled at 44k1. If there's interest I'll design one for 88k2 too.

About the PCB - so far no problems. Its of average quality, not as good as that in the DAC-AH but that one's exceptional. I've pulled out a few capacitors (lazily too, without using a solder sucker, just heat and jiggle) so far without any issues whatsoever so I think you'll be fine.

@matt - yes you're right, the PCB's quite simple, except for having island groundfills where most have an overall groundfill with the odd slot in. If anyone wants to go ahead and design a 'proper' PCB then start a thread and I'll certainly be a crowd member there :) As for your sourcing question - nope, no idea about any sellers beyond China. Guess patience is the word :p
 
Last edited:
I've just put a dent in one remaining weakness of this DAC - the bass. Its a very simple mod to add capacitance to the TDA1387's supply. I soldered some 1.5mm copper wire to the two pads for the lytic (just beneath the chips in Matthieu's pic) and used that wire as 'bus-bar' to solder across 4 * 15,000uF 6.3V Panasonic NHG caps in a horizontal position. This tightens up the bass nicely. While I was fiddling I couldn't resist tweaking up the supply voltage slightly, to almost 6V (the max for the DAC chips).
 
Tried it stock today (LT1028 in my case). Better than I've expected as-is, not as harsh and strident as was a Ha-Info based one (and more exepensive AFAIR), yes a bit nasal and not that smooth on highs, but very listenable already. Can't wait to hear it finished even if I'll go for a bit less capacitance in the end. In stock form on our set-up its main quality was it's joyfull presentation.
I'm lazy (and ill!) so I'll need to be spoon fed ;) What value of resistor you changed at the TL431 Richard? Maybe safty (caps...) calls for 5.5/5.8V? Thanks!
 
Last edited:
The TL431 I was lazy and put another resistor in parallel with the existing 10k - it was 13k because I happened to have a reel of them nearby. Then I added another 13k a bit later in parallel with those two. You could just slap down a single 6k8 on top of the 10k already there. Oh perhaps you're meaning the DAC's TL431 haha - let me see I also paralleled another resistor there (3k is fitted in mine) and I needed a value of about 2k2 (for 5.9V) so I put on top of the 3k a 9k1 (erring on the side of safety). Note the resistor here is the (logically) lower one - the one connected to 0V, that's on the top of the PCB I think. The upper side resistor is below.

About caps and their working voltage - a few years back I got bitten with some Taobao caps not being the advertised voltage in reality and some blew up. So now I test all caps for leakage at above their rated voltage before I use them. These Panasonics I've put on the supply went well above 6.3V very consistently (actually to 9V before leakage climbed significantly).

If you were only to do one mod, the most significant one I'd say for listening pleasure is deleting those feedback caps (2n2s) around the opamps. That kills a lot of the harshness and improves the depth. After that upgrade the opamp supply caps with ceramics and bias the output with a resistor to -VEE (I'm using 2k2 with rails about 8V). Then change the pin7 caps on the DACs which kills most of the nasal quality and improves the LF clarity.
 
Now do something interesting with this project, folks: OVERSAMPLE ;)
You can get use SAA7220 or SM5840 or DF1704 or (if you were "smart" 'nough to stock up on 'em) PMD100 or PMD200. If you're real good, you may be able to OS in software or firmware/DSP (ala Wadia, Theta, Meridian).
 
Because...

Please explain why oversampling is interesting?
Because I've tried NOS in many flavors and colors ... and found it ALMOST universally boring.
I don't think I'm in the minority (which in and of itself is unimportant): digital audio engineers began using OS (roughly starting in 1984, with 2x) to address the analog brick-wall issue. What they didn't do -- at least the mainstream manufs. -- was supply the OS chip with clean pwr (via its own regulator and plenty of decoupling). "Audiophile" modders (Meridian, Mission, Mod Squad ,etc.) took basic Philips or Sony CDPs and addressed these issues.
Today all "high-end" manufs (and device manufs.: ESS, TI, Cirrus, Wolfson) use this topology (OS) almost exclusively.
Sure you can polish and tweak AROUND a NOS design (DIY audio is full of those projects!) -- and you'll get better (general) sound than OS without these tweaks/mods (see my prev post on diyparadise Monica/Mojo) or your own (??)...
1465d1420014813-new-year-new-dac-p1010142.jpg


... so ...
with all that DIY enthusiasm and plenty o' time-to-kill ... would sticking a simple OS/DF into the I2S line (reporting your findings in this forum) really be that hard?
Come on ... where the sport? ;)
 
Last edited:
Because I've tried NOS in many flavors and colors ... and found it ALMOST universally boring.

I really, really wanted to like oversampling because it would have given me a great excuse to include a CPU in my DAC designs (which few people have). But guided by my ears I found the faster I ran my DAC, the less interesting the sound. So my experience is diametrically opposed to yours.

Seeing as my digital filter was a software one it was dead simple to compare 'bypassed' (i.e NOS) with '2X OS' and '4X OS'. The ARM CPU I used needed around 10mA I seem to recall which is a lower power requirement than the filter chips I've encountered. I also tried a pure hardware oversampling approach to 2X OS.

... so ...
with all that DIY enthusiasm and plenty o' time-to-kill ... would sticking a simple OS/DF into the I2S line (reporting your findings in this forum) really be that hard?
Been there, done that and no it wasn't that hard, a very useful learning experience.
 
Oversampling redux

Of all the classic Red Book designs I've tried, Naim's TDA1541A/SAA7220 is the one I liked best. There's a reason why they located the 7210/7220/1541 I2S tracelines super close. ..
cds1.jpg~original
.
And this is one reason I think OS in software is compromised. Another reason is that a purpose-built DF (incl. FPGA/DS firmware as well as the filter section of modern DAC ICs like PCM1792) are better than some software running in modern (=generic, consumer-grade) PC/tablet/etc.
In the manuf industry, where I sometimes work, purpose-built devices are often used because of performance/speed. Although any modern Intel/AMD CPU processor is faster, pre se, than a given ad hoc processor or FPGA, the thing that counts in the end is application specificity. You'd be surprised at what military/NASA continues to use (=legacy ,old-school) because that's what STILL works best for the app.
As far as NOS ... warm and tubey for sure. And it has detailed upper bass (surprising) that (IME) is superior to OS. But WRT punchiness/slam/detail, OS wins IMB.