lingDAC - cost effective RBCD multibit DAC design

Frequency response in simulation

Here's what the filter looks like in LTSpice - you'll note there's almost 3dB lift at 18kHz to counteract what's commonly called the 'NOS droop'. Beyond 18kHz the response falls fairly steeply to a rejection of around -44dB above 30kHz. This doesn't eliminate all images from the DAC - to do that with NOS at the 16bit level would call for a filter more than twice as complex as this one.

The stop band attenuation at higher frequencies (>100kHz) is greatly helped by the series resistors (R9,10 etc) at the expense of reduced attenuation at the zeroes, a price well worth paying I feel.
 

Attachments

  • lingDACFR.png
    lingDACFR.png
    4.7 KB · Views: 1,390
Disabled Account
Joined 2015
modification proposition

The filter has a resonant frequency of about 17khz , this is fundamental error . The frequency should be half of sampled frequency that is 22.4khz (24) for 48khz . I modified for you with the same damping factor but with flatter response up to 20khz . What is marvelous with Z filter as this how Dual (German brand) has baptized it, it's phase response . The filter has a constant delay of 19us . If you compare the phase shift with a delayed signal ,the phase error at 10khz is+1.4° from 1khz as references and -7.4°at 20Khz . In other words the phase shift throughout the whole spectrum is+/-4.5° only . Now after 30 years I finally understood why Dual non over sampled models as cd1045 had so high presence sound .
Bravo
 

Attachments

  • NONAME.JPG
    NONAME.JPG
    107 KB · Views: 1,774
  • NONAME2.JPG
    NONAME2.JPG
    55.9 KB · Views: 1,162
Looks like you increased the cut-off frequency (and consequently the stop-band start frequency) and got rid of the NOS boost. I can't see why this is going to be better though - you reckon the phase response is the most critical thing about a filter? Without the NOS boost you'd need to run the DAC oversampled to get a flat passband, maybe 4X OS?
 
Here's what the filter looks like in LTSpice - you'll note there's almost 3dB lift at 18kHz to counteract what's commonly called the 'NOS droop'. Beyond 18kHz the response falls fairly steeply to a rejection of around -44dB above 30kHz. This doesn't eliminate all images from the DAC - to do that with NOS at the 16bit level would call for a filter more than twice as complex as this one.

The stop band attenuation at higher frequencies (>100kHz) is greatly helped by the series resistors (R9,10 etc) at the expense of reduced attenuation at the zeroes, a price well worth paying I feel.

Ahaa... nice to know: I was wondering in fact if these filters were also compensating for the NOS droop !
While we wait for all schematics and since I have some TDA1543 dacs waiting for a filter: could I wire this filter to a TDA1543 output as is or would it need some modifications ? Thanks B.
 
There's a crucial difference between the TDA1387 and TDA1543 in relation to the filter-I/V stage and that's their output compliance voltage ranges. The 1387 can source current down to 0V on its output whereas the 1543 can only do so down to 1.8V. Since the filter-I/V board applies about 1.1V to the DAC's output, it won't work un-modified with the 1543. Fortunately there's a simple fix - just change LED4 (currently a red) to something with a forward voltage higher than 2.5V - a blue one should do the trick. To maintain the supply voltage at 9V the rest of the LED string will need changing so as to keep the same 6.9V reference voltage.

There will be another couple of minor mods - the output swing of the 1543 is 2.3mA so the I/V (R6,41) resistors will need reducing (4k7 down to 2k4) and the DC offset trim will need attention.
 
Last edited:
There's a crucial difference between the TDA1387 and TDA1543 in relation to the filter-I/V stage and that's their output compliance voltage ranges. The 1387 can source current down to 0V on its output whereas the 1543 can only do so down to 1.8V. Since the filter-I/V board applies about 1.1V to the DAC's output, it won't work un-modified with the 1543. Fortunately there's a simple fix - just change LED4 (currently a red) to something with a forward voltage higher than 2.5V - a blue one should do the trick. To maintain the supply voltage at 9V the rest of the LED string will need changing so as to keep the same 6.9V reference voltage.

There will be another couple of minor mods - the output swing of the 1543 is 2.3mA so the I/V (R6,41) resistors will need reducing (4k7 down to 2k4) and the DC offset trim will need attention.


Thanks for the info !
B.
 
SE classA output buffer board

The third board in the lingDAC set is designed to buffer the relatively high impedance output of the filter-I/V board so its suitable for being fed to interconnecting cables, or even to moderately sensitive headphones.

I'm using a relatively new Toshiba MOSFET design chosen for its high transconductance, lowish capacitance and reasonable power dissipation (while still in a SOT-23 package) as the output driver. The choice of a P-channel rather than the customary N-channel part reflects the desire to keep the load-induced power supply noise as low as possible.

Whilst the schematic shows 2,200uF output coupling caps, these are the solution for headphone use. For line level the caps can be much lower value (4.7uF to 10uF) film caps. R10 & R18 will need increasing to 47k or higher when film caps are used.
 

Attachments

  • Schematics 0DAC-SEBUFF.pdf
    28 KB · Views: 525
I'm excited to see this thread! I have been lucky to have Richard's LingDAC 1.1 (4x chip variant) in my system for a couple months now. It sounds very good.

There has been some discussion about the analog filter and the phase/frequency response tradeoff. I can say without hesitation that Richard's filter brings the 1387 to life in a remarkable way. It is easy enough to bypass the filter board for A/B listening back and forth. I did this, and the sonic differences were dramatic enough to make it a "no-brainer" to leave the filter engaged permanently.

Basically the filter obliterates the slow, dark, soft coloration intrinsic to the unfiltered 1387 sound, while preserving the things that make the chip special in the first place--the top-to-bottom cohesiveness, timbral definition, and full-bodied openness. Instruments with HF content, especially cymbals, are rendered with far more transient speed and realism. In our conversation, Richard summarized the effect as bringing the music "into focus," and this is an apt description. After my previous experience with Matt's sublimely musical and mellow 1387 Pi hat (unfiltered single chip), I am now really amazed at how detailed and precise (dare I say "reference level?") the 1387 can be.

How much of this magic comes from the treble droop correction, and how much from filtering out the ultrasonic image, I can't say. It would indeed be interesting to hear it both ways. Re: Kokoriantz's Dual filter, I am convinced from my past experiments with digital EQ that phase linearity can benefit imaging and transparency. Surely there is an optimal compromise we must find with our ears. :) Component quality of course matters too...I would eventually like to try building up a DAC filter using all PP caps, for instance.

Here is a pic of my test setup. Source is Volumio on RPi stacked with an IanCanada IsolatorPi and Allo Kali. Each board is fed by its own modified, grounded 5V Jameco linear wall wart. LingDAC gets 8.1V from a similar Jameco 12V supply, stepped down through an eBay LT3042 module, followed by a CRCRCRC filter. I am using 6.8uF ClarityCaps CSA to couple LingDAC to my modified Topping Tp20 mk2 Tripath amp. Bookshelf speakers are Axiom M3.
 

Attachments

  • IMG_3303.jpg
    IMG_3303.jpg
    957.6 KB · Views: 1,084
Last edited:
Got my lingdac fired up over the weekend, so was able to finally listen to it.

I plan to post some impressions in a couple weeks or so, want to let it run in a bit.
Then I'll give it a good listen, and then move my source over to my modded soekris dac and see how it compares.

I'm powering my dac with three lifepo4 batteries, which goes through a couple diodes the Abax added to drop down the voltage to around 8.5VDC.

Source is a SD Card player that I reclock with a potato semi flip flop just before the DAC using NDK clocks. The SDP and FF and clocks are powered from another lifepo4 battery.

My quick impressions is that I was very impressed with this dac, and I could live with the dac long term. I think the SEBuff stage gives this good dynamics, and it has plenty of detail.

Would like to build a Ver 2 DAC at some point, and see how that sounds, the bass in the V1 could use a little help.

I noticed my SEBuff board has a few ceramic caps in parallel, in series with a 10 ohm resistor. I'll probably change out the ceramics with a 10uf film cap, and see how that sounds.

The boards are surface mount, but don't look too bad to build. A magnifier and a fine tip help a lot with small parts.

The part I think would be a challenge to build this dac is the filter, and getting the LC values right. If this part can be figured out, the rest looks doable.

Randy
 
Would like to build a Ver 2 DAC at some point, and see how that sounds, the bass in the V1 could use a little help.

I noticed my SEBuff board has a few ceramic caps in parallel, in series with a 10 ohm resistor. I'll probably change out the ceramics with a 10uf film cap, and see how that sounds.

Interesting. In my setup, bass sounds on par with my other sources. I agree that dynamics are good with the buffer.

I removed the ceramic output caps and resistor, and replaced with 6.8uF films. See pic in post 50.

The big purple PP caps dwarf the board rather comically, and I will eventually just short the board terminals and wire the caps to RCA jacks when I case this thing up.

But if you want to keep with the compact theme, I also tried 6.8uF WIMA MKS, which fits directly on the board and is the largest value WIMA film cap available with 5mm lead spacing, I believe. These polyesters are slightly less transparent than PP but still a huge upgrade IMO.
 
I don't have a size problem, here's my setup mounted on a block of wood.
I have some 10uf film caps I like that I'll use here, and see how it sounds.

First impression was that bass was better with my soekris, but as I said I'm not going to do that direct comparison for a couple of weeks.
Need to change out the caps first, and then give it some time to run in.

Randy
 

Attachments

  • 20180721_100936.jpg
    20180721_100936.jpg
    1,001.6 KB · Views: 827
Gerbers for SE buffer

PCB data attached for the 3rd lingDAC board - which incidentally can be used as a stand-alone headphone buffer for applications which require no voltage gain. For example attached to a mobile phone and powered from the OTG USB voltage rails.

As @koko has already noted, the output impedance when using the J327s at a few tens of mA bias is pretty respectable for a single transistor single-ended buffer.:)
 

Attachments

  • Gerber_SEbuff.zip
    35.7 KB · Views: 190