I've been comparing Kubelik DAC connected to digital out of a Sony S900ES vs the internal DAC of S900ES. There is no gross differences. At times it was hard to tell the difference. There is a very minor difference, just no sure how to nail it down yet.. Kubelik has improved my TV sound immensely along with music server.
Having done some comparative listening between Celibidache and Kubelik I found that the higher frequencies were quite a lot more 'clouded' on Kubelik vs the more expensive DAC and I've been intensely curious to find out why.
This derivative of Kubelik called 'Abbado' is an experiment to uncover the reasons for Kubelik's reduced clarity. Its taking a rather unconventional approach of running at a much higher output level (6VRMS) rather than the more usual 2V. The reason for the higher output voltage is its a way to gain more SNR. Kubelik uses a noisier opamp (OPA2209) than does Celibidache so having more signal is a way to 'level the playing field' so to speak. Because of the extremely unconventional output level I'm not sure if this DAC has a commercial appeal but it dovetails quite nicely with a revised 'monoAMP' with lower gain : https://www.diyaudio.com/community/threads/transformer-input-tda8932-mono-amp.357678/post-6733863
With its 6V output I mean to play it straight into my 600ohm DT880s. For that purpose it has a much higher output current capability than Kubelik (SOT89 transistors in place of SOT23s).
This derivative of Kubelik called 'Abbado' is an experiment to uncover the reasons for Kubelik's reduced clarity. Its taking a rather unconventional approach of running at a much higher output level (6VRMS) rather than the more usual 2V. The reason for the higher output voltage is its a way to gain more SNR. Kubelik uses a noisier opamp (OPA2209) than does Celibidache so having more signal is a way to 'level the playing field' so to speak. Because of the extremely unconventional output level I'm not sure if this DAC has a commercial appeal but it dovetails quite nicely with a revised 'monoAMP' with lower gain : https://www.diyaudio.com/community/threads/transformer-input-tda8932-mono-amp.357678/post-6733863
With its 6V output I mean to play it straight into my 600ohm DT880s. For that purpose it has a much higher output current capability than Kubelik (SOT89 transistors in place of SOT23s).
I've now built 4 'Abbado' DACs, the first two with 6 TDA1387s. On the third one I wanted more chips so I stacked them two deep. The drawback with more chips is more heat in the 'dropper' resistors - this required an extra small PCB for those resistors, to accommodate twice as many as will fit on the PCB. Since dealing with all that excess heat is a bit of a nuisance I was curious to try TDA1545A as that chip will output twice as much current as a single TDA1387. So in theory I could get the same SNR but with just six chips and have less heat as the supply current is a bit lower. What's not to like?
Actually there's one thing not to like - the 1545 isn't I2S input, but I handled that with the CS8412 on the PCM56 DAC I bought recently, it was already configured in the correct output mode. I had high hopes on firing up the TDA1545 version of Abbado but in reality I was greeted with music drowned in a wash of crackling. Unlistenable - it sounded like one or more chips was broken. Turning them off one by one clearly demonstrated they all contributed to the crackling. So unless my whole tube of chips was suspect, that wasn't going to cut it as an explanation.
To cut a long story short - it turns out the DS isn't comprehensive enough when running the TDA1545 with higher than 1mA output current. This explains why I got strange results with my experiments with this chip going back almost 10 years. TDA1545 probably can sound as good as TDA1387 I reckon once the weakness of the DS is understood and circumvented. However its early days, more listening is called for.
Actually there's one thing not to like - the 1545 isn't I2S input, but I handled that with the CS8412 on the PCM56 DAC I bought recently, it was already configured in the correct output mode. I had high hopes on firing up the TDA1545 version of Abbado but in reality I was greeted with music drowned in a wash of crackling. Unlistenable - it sounded like one or more chips was broken. Turning them off one by one clearly demonstrated they all contributed to the crackling. So unless my whole tube of chips was suspect, that wasn't going to cut it as an explanation.
To cut a long story short - it turns out the DS isn't comprehensive enough when running the TDA1545 with higher than 1mA output current. This explains why I got strange results with my experiments with this chip going back almost 10 years. TDA1545 probably can sound as good as TDA1387 I reckon once the weakness of the DS is understood and circumvented. However its early days, more listening is called for.
I see from the date on the last post that I've not posted here for over a month! Just got in some 'Abbado r2' boards which have been updated to use TDA1545AT in place of TDA1387Ts. The 1545 needs a small number of external components to feed its Iref (pin7) input and I've been playing around with power supply voltage and output compliance voltage vs the Iref input current. A couple of interesting findings :
1) 2mA output current isn't available at all supply voltages, looks like something in the region of 4V is the minimum required to get 2mA.
2) By setting the output compliance voltage somewhat higher than on the DS the internal current source can be disabled which holds an advantage for supply current - setting the output compliance to the positive supply allows the DAC to deliver 2mA with lower supply current than the 1387 takes for delivering 1mA. This improved efficiency is a significant advantage as Kubelik's design was limited in practice by the heat dissipated in the dropping resistors to the DAC supply. I won't be using the same 2N7002 cascoding MOSFET though as the output compliance voltage needs to be held with a tight tolerance - even a couple of hundred mV turns on the internal current source to a degree. Fortunately a Rohm MOSFET saves the day here - its the FET with the lowest gate threshold voltage I've ever seen, similar to the Vbe of a bipolar transistor. Its tolerance so far seems very tight in practice.
1) 2mA output current isn't available at all supply voltages, looks like something in the region of 4V is the minimum required to get 2mA.
2) By setting the output compliance voltage somewhat higher than on the DS the internal current source can be disabled which holds an advantage for supply current - setting the output compliance to the positive supply allows the DAC to deliver 2mA with lower supply current than the 1387 takes for delivering 1mA. This improved efficiency is a significant advantage as Kubelik's design was limited in practice by the heat dissipated in the dropping resistors to the DAC supply. I won't be using the same 2N7002 cascoding MOSFET though as the output compliance voltage needs to be held with a tight tolerance - even a couple of hundred mV turns on the internal current source to a degree. Fortunately a Rohm MOSFET saves the day here - its the FET with the lowest gate threshold voltage I've ever seen, similar to the Vbe of a bipolar transistor. Its tolerance so far seems very tight in practice.
Nice find, congrats!
May I ask in practice how does all this translates into sonic advantages and/or price reduction, as I guess these are really the key features of Kubelik?
Thanks and congrats again for your relentless quest!
Claude
May I ask in practice how does all this translates into sonic advantages and/or price reduction, as I guess these are really the key features of Kubelik?
Thanks and congrats again for your relentless quest!
Claude
If you'd like more technical background, I recently updated my Hackaday pages - https://hackaday.io/project/27001/log/204631-moving-beyond-kubelik
Definitely Abbado sounds more dynamic than Kubelik, not just because of the 5th order (vs 3rd order) filter, also seemingly because I'm no longer using a transimpedance stage for doing I/V. I'm back to traditional passive I/V again after all these years as it gives superior noise characteristics. The extra inductors in the filter mean Abbado won't be cheaper than Kubelik but the increase in SQ is definitely worth a fistful of extra dollars.
Definitely Abbado sounds more dynamic than Kubelik, not just because of the 5th order (vs 3rd order) filter, also seemingly because I'm no longer using a transimpedance stage for doing I/V. I'm back to traditional passive I/V again after all these years as it gives superior noise characteristics. The extra inductors in the filter mean Abbado won't be cheaper than Kubelik but the increase in SQ is definitely worth a fistful of extra dollars.
Interesting read, thanks!
So I guess we are likely to see Kubelik replaced soon by Abbado, once finalised?
So I guess we are likely to see Kubelik replaced soon by Abbado, once finalised?
I'm not sure as Abbado doesn't lend itself to being turned into kits very well, given its need for hand-matched capacitors in the filter. At present I'm planning that its only going to be available ready-built.
There is also the issue of the different digital format to be dealt with. It doesn't take I2S - this is fine for people who only want S/PDIF and Toslink inputs as the S/PDIF receiver chips used handle the conversion. USB input cards though do not, they only output I2S to my knowledge.
There is also the issue of the different digital format to be dealt with. It doesn't take I2S - this is fine for people who only want S/PDIF and Toslink inputs as the S/PDIF receiver chips used handle the conversion. USB input cards though do not, they only output I2S to my knowledge.
Last edited:
Richard,
even if you know all of this eventually, I want to mention: it shouldn't be too difficult to convert i2s into eiaj (Sony or japanese) format. Yeo from DIYparadise once did if for his Monica3 dac. At this time he offered a small board for that, yet he has been out of business for a long time. This pic is from his still existing website.
As a second option the JLsound USB2i2s board is able to output RJ16 format, which should be the right one. The board is really a good one, yet not cheap.
Cheers, Ernst
even if you know all of this eventually, I want to mention: it shouldn't be too difficult to convert i2s into eiaj (Sony or japanese) format. Yeo from DIYparadise once did if for his Monica3 dac. At this time he offered a small board for that, yet he has been out of business for a long time. This pic is from his still existing website.
As a second option the JLsound USB2i2s board is able to output RJ16 format, which should be the right one. The board is really a good one, yet not cheap.
Cheers, Ernst
Thanks @ernesternest - yes I already have a circuit to do the I2S->EIAJ conversion, it'll be interesting to compare that with the board you're showing. Monica DAC was another example of a TDA1545A design I recall from years ago. I think you're right that RJ16 (assuming that means right-justified 16bit) would be appropriate. That JLsounds is indeed a USB interface that handles non-I2S output formats but it does cost quite a lot more than the run-of-the-mill CM6631A offerings.
Thinking out loud I'm wondering if Abbado should become a balanced output (hence mono) DAC requiring a digital interface card to drive it. Gotta do some listening to see if balanced brings any advantage to SQ, it neatly solves a circuit design issue that's been bugging me.
Thinking out loud I'm wondering if Abbado should become a balanced output (hence mono) DAC requiring a digital interface card to drive it. Gotta do some listening to see if balanced brings any advantage to SQ, it neatly solves a circuit design issue that's been bugging me.
Here's a thread on EEVblog which sheds a little more light on the I2S->EIAJ conversion board above. I wanted to find out which ICs were being used and it turns out they're 74HC174 which are hex flip-flops. With 3 hex chips that gives up to 18 bit-clock delays.
https://www.eevblog.com/forum/projects/i2s-to-eiaj-glue-logic-digital-audio-converter-help-needed!/
My schematic uses 3 chips but I'm using 74HC164 shift registers each with 8 blt-clock delays. With a 32bit data frame I2S format (the normally encountered variety, with f(BCK) = 64*f(LRCK)) I need 15 bit-clock delays of the data field. Plus inversion of LRCK.
https://www.eevblog.com/forum/projects/i2s-to-eiaj-glue-logic-digital-audio-converter-help-needed!/
My schematic uses 3 chips but I'm using 74HC164 shift registers each with 8 blt-clock delays. With a 32bit data frame I2S format (the normally encountered variety, with f(BCK) = 64*f(LRCK)) I need 15 bit-clock delays of the data field. Plus inversion of LRCK.
Richard, I do have an old Monica USB dac, I think it is Monica3. I can send you some pics, if interested. The board is not black, so traces are good to see.
Hi Richard, this is a pic of the Monica3 DAC. It shows how the shift registers are connecte. The colored points are connected via traces on the other side of the board - via vias. Hope it helps. BR, Ernst
Today I finished building up two balanced Abbados along with the digital paraphernalia needed to split the I2S into two streams, L +/- and R +/-. Pleased to report that its sounding rather promising straight away, a noticeable bit more 3D and detached from speakers than when run single ended. Even wifey said she noticed from the next room (couldn't resist the audiophile trope 😛).
Heat is still an issue for this design as the method of disabling the internal CCS turned out to introduce quite a bit of 2nd and 3rd harmonic distortion, enough to be visible on my 'scope (means considerably above 0.1%). And this didn't get the thumbs up SQ-wise from wifey. So still engaged in playing with ways to get lower current consumption while keeping the distortion in check.
Here's a plot of two things in relation to the 5th order filter that Abbado is using. The filter's magnitude response is shown in purple - its a Chebyshev with about 1dB peak-peak ripple. The green plot is indicative of the Zout of this filter, its made by applying a current source to the output of the terminated filter with the input open-circuit. The Zout is important because the noise of the passive I/V stage follows this curve - I shall double-check later with an LTSpice noise analysis.
The filter's been designed to give a minimum in the output impedance at the frequency around which the ear is most sensitive to noise. I figure that's around 6.3kHz and I've taken that number from the peak of the ITU-R468 noise-weighting curve.
The ITU-R468 curve is shown here : https://en.wikipedia.org/wiki/ITU-R_468_noise_weighting#/media/File:Lindos3.svg
The filter's been designed to give a minimum in the output impedance at the frequency around which the ear is most sensitive to noise. I figure that's around 6.3kHz and I've taken that number from the peak of the ITU-R468 noise-weighting curve.
The ITU-R468 curve is shown here : https://en.wikipedia.org/wiki/ITU-R_468_noise_weighting#/media/File:Lindos3.svg
Here is the PCB for the 'Balancer' - what this does is it takes in a stereo I2S stream and splits it into two mono streams, each mono stream has positive and negative polarities so it'll turn a pair of stereo DACs into balanced dual mono DACs.
There's one significant limitation at present - this will only work at a BCK of 2.8MHz, no higher. That's because U5 is a very slow 4000 series CMOS chip that has no equivalent in the HC range. So 44k1/32bit I2S or 88k2/16bit are all it can cope with. That's not a significant limitation I have since my music is pretty much all 44k1.
There's one significant limitation at present - this will only work at a BCK of 2.8MHz, no higher. That's because U5 is a very slow 4000 series CMOS chip that has no equivalent in the HC range. So 44k1/32bit I2S or 88k2/16bit are all it can cope with. That's not a significant limitation I have since my music is pretty much all 44k1.