Project : The non oversampling dac

Its been a long time that I have been looking for a do it yourself dac. Until now I used a Sony ES505 and the best Pioneer cd player the PDS 06 and its successor the PDS 707 ( I still have 2 for sale ). Both are excellent value for money, and having tried many outboard digital analog converters on both machines from Sonic Frontiers ect, I was never really under the impression that they improved the sound.So I never got further then the stage of gathering information.Some clients of mine have Wadia equipment and those didnít impress either.Then while surfing on the net, I stumble on Galavottiís home page and I read about his project of building a non oversampling dac. Also there is a link to Kusonoki theory article that appeared in a Japanese magazine known as MJ. I was very impressed, this is the first time that I felt real progress has been made.The web also showed similar designs by Thorsten Loesch and Andrea Ciuoffoli and Jutta Tolonen. All very similar but with their own personal touch . A bit of research and a month later, I draw schematic v1 and approach a few customers about this project and they like it, since my last project was with Stephane (the Outer Limit ) I give him a call and ask him to look the project through as he has lots of experience with digital circuits and pcb design for critical RF applications . The revised power supply is all courtesy of him, he managed to use just one transformer for all (it does have 5 different windings)And it is also fully capable of supplying current to a Philips vam 1254 high end drive mechanism which will be added later ( meaning for part 2 )We added a CS8412cp for the people who wish to use it as a stand alone dac.However the final version of this dac uses 2 TDA1541A as this seems to sound the best in conjunction with the Philips CD player unit . For test we also tried the on board DAC on the Philips VAM-1254, but that was not bad compared to the Pioneer but not on par with the non oversampling.The circuit Very straight forward :All power supplies have current sources for optimal shielding against each other, 2 ground planes are used : digital and analog.Elna cerafine elcos used, everywhere.Also a pulse input transformer from Lundahl for optimum signal transmission quality.Followed by the CS8412, this all goes straight to the TDA1541 from Philips which still is the reference 16 bit dac, there the signal is sampled in top quality mkp capacitors matched to 1%. ( whether its important or not I still donít know )Then a 470pf silver mica is used for the clock .In the analog output stageThe problem of using a filter with almost no phase shift at 20Khz, makes it quite impossible a task, many solutions tried and failed.We thought we could succeed but !) Here lies the heart of 1 of the problems of the non ovesampling Dac.What is wrong :Please note that no one mentions thisProbably due to the human nature, of not liking to admit failure!!!Stephan, decides to put the dac on the test bench to see how it performs because he suspects something flawed and its always nice to see that the specs are not too bad.1Khz sine wave is good but you clearly see that the sine wave is made of steps, the 4 Khz sine wave is quite flawed with fewer steps making the sine wave and 20Khz is catastrophic. 2 steps represent the whole sine wave, in comparison my pioneer PDS06 has a perfect sine wave with no steps even at 20KHz.However this is not the real problem as it is the nature of the non oversampling non filtered design.The following problem is worse :

Study of this phenomenon, shows that the clock at 44,1Khz has sub carriers, these are of harmonic nature and therefore when used in conjunction with tube amps they inter modulate and the result on the speakers is terrifying!!! 20Khz test tone = first peak24Khz = first byproduct†† (44Khz-20Khz)

64Khz = 2nd byproduct†††† (44Khz + 20Khz)

68Khz = 3rd by product†††† (44Khz+24Khz)

this is measured on the output of the dac with a ref cd playing as frequency generator.Put in words : The whole audio band is duplicated around the carrier of 44,1KHz, above and beneath and this continues with every harmonic = 88.2Khz ectso the problem is the entire audio band 0-20Khz is reproduced thus from 44Khz + 20Khz = 64Khz and 44Khz-20Khz = 24Khz this particularly hurts audio reproduction as they inter modulate so we get if we have a 20Khz sine wave in the dac we get a duplication at 24Khz but worse we have the same energy as an inter modulation at 4Khz !!!!!!!!Because the tube amps will inter modulate 24Khz-20Khz.Very easy to test we take a ref CD with test tones and listen to them and indeed at 16Khz you can clearly hear a lower frequency through and at 20Khz, barely audible this frequency, we crank up the volume of the pre amp and the 4Khz tone appears very nicely.So basically this is a disaster: not even something to publish.However listening tests suggest the problem is not so bad, as the sound of this dac is practically identical with what my Pioneer ( 20bits 8xoversamplig BB dac and legato link filtering ) produces, actually it is very difficult to hear a difference.Which is the reason why this project almost ended up on the shelves, however having a few customers already involved meant it had to be developed to the end.That is with the dacused as dac and withoutits own transport, the result is far better with the Philips module and the use of the I2C bus.The I2C bus seems to cure many of the problems that are probably Jitter related but make a world of difference.

The other major step forward is the use of paralleling TDA1541, I would describe this as major upgrade, most noticeable is the stability of the whole image when many instruments are playing at the same time, the stage says wide and all instruments stay focused otherwise it gets quit blurred. Solution:Yes, there is a solution. Well since filtering didnít work as anticipated we had to look for another solution, and Stephane had an idea oversamplingWell not that type of over sampling but another one, as the principle of the dac is non over sampling and the ideas behind it are sound and healthy we stayed in this but slightly changed it.What changed?We call it a bit doubler, basically every audio bit is doubled and effectively this puts the clock at 88Khz and keeps the idea as the chip is not doing any over sampling or error correction.The sacred principle behind this dac is exactly that and its respected.The gained advantage is now we can simply and gently filter and obtain a simple filter.The output signal measures better and sounds better.

Here you see the 20Khz test tone and the first byproduct is at 68Khz (88Khz-20Khz) and the 2nd byproduct at 108Khz (88Khz+20Khz)and the 3rd at 146Khz .

They are already attenuated by a gentle 3rd order Bessel slope and will not pose any problems in the audio band.The current to voltage conversion is done by simple dale 1w 68 ohm resistors , 2 in parallel to avoid clipping of the dac output chip. ( Overloading is the number 1 common source of distortion in audio DACís )This is the one most common place of sound coloration, if the value of the resistor is to low then the output and dynamics are low, if the value is to high then dynamics are better but distortion occurs as the TDA cannot supply the requested current without some type of soft clipping , this causes heavy sound coloration.The analog signal is amplified by a tube amp : a 5842 with current sources as to have optimum quality, this is almost as good as anode chokes. The Ķ=43 so with the current sources we can say a gain of 40x approx .We also tried 2 current sources with boot strapping but it didnít sound better, it sounded worse.And a relay mutes the tube stage when it is powered up.The tubes are biased with negative bias, I just canít tell you guys enough how much better this is.The power supplyís :Pcb A : >everywhere 3A regulators are used : LM1085 and LM337 , one source 7vac winding does the job , voltage is doubled and doubled again when a higher voltage is needed , the multiple regulators filter out all riplle and most importantly isolate the supplies in ac from each other . for the dac and input receiver we need :57mA 2times

+5v††† 83mA

-5V††† 83mA

-15v†† 83mA


Pcb B:

Power supply for Philips cd vam 1252:

Here we use a separate toroid of 2x7vac again regulators LM1085 3A

A voltage doubler is used to obtain the Ė20v for the gas light display bias voltage

The voltages needed are :

+9v††† 1A

+5v††† 1A


2,5vac used for the display of the Philips (actually 2vac is more then enough)

Pcb C:

The high voltage for the tube output stage is rectified here, a triac is used as delay switch. The high voltage is mos regulated at 220 volts dc

The low voltage for the filaments and for the relais and the triac firing are all obtained here and through a 4060 counter activated.


Pcb D: the dac itself V1,0Also pictures of the almost finished Outerlimit cd player .

To be continued

copyright Benny Glass 2003