Building the ultimate NOS DAC using TDA1541A

Hi,

I just joined diyAudio.com. I have spend my entire life designing, repairing and buildingjavascript:smilie(':smash:') electronic equipment, electronics is my passion. I have designed a lot of High-End audio equipment over the last decades. The following projects are the "newest":

-Fully symmetrical cascode MOSFET amplifier with high resolution microcontroller based VU meters.
-Modular 6 channel control amplifier with remote control
-Modular input switchbox
-Twin TDA1541A DAC in differential mode with 8th order active hybrid filter
-Sonic resonators (passive semi 4-way 360 degree omnidirectional radiators
with real-time error correction). How do they sound? hearing is believing!
-Non inductive copper wire resistors for High-End passive filters, using stranded wire and a spiderweb winding technique. These resestors have extremely low noise and improve sound clarity.

javascript:smilie(':att\'n:')I am now developing a NON oversampling DAC using 4 X TDA1541A. The 44.1 kHz problem (ultrasonic interference in the audio band) is solved by using a trick I named Direct Interpolation (well this system had to have a namejavascript:smilie(':D')). This results in a virtual sample-rate of 176.4 KHz, 18 bits resolution, quadruple output voltage, improved signal to noise ratio and lower distortionjavascript:smilie(':)'). I used a second order bessel filter at 40KHz for linear group delay just in case. The sound and dynamics are breathtakingjavascript:smilie(':bigeyes:') since both phase-shift and ultrasonic interference are greatly reduced. Now finally I could hear music the way it was originally recorded. At the moment I am optimizing the Direct Interpolation system. Tips or suggestions from you all are highly appreciated


- Why the TDA1541A? it uses current sources based on dynamic element matching instead of resistor networks. The current sources are externally decoupled so they produce clean signals. Furthermore a smart design elliminates signal transients all together. In my humble opinion this is the purest form of D/A conversion, and the sound quality of these DAC's prove this. It is really a shame this state of the art DAC is discontinued, but I guess that is called "progress". Modern DAC's have the digital brickwall filter and decimation noise generators built in. I cannot follow the logic of spoiling a fine signal by decimating and then try to filter it out the mess you've createdjavascript:smilie(':xeye:'). The high oversampling frequencies will make sure there is lots of addittional noise and interference to listen tojavascript:smilie(':('), it's as if you put your High-End amplifier inside a PCjavascript:smilie(':eek:').

- Why more than 1 DAC? With 2 ore more DAC's linearity errors are reduced, and normal lower cost TDA1541A's can be used. Signal to noise ratio improves, differential output can be used, elliminating residual interference and... the DC component can be cancelled out, enabeling a fully DC coupled DAC. In the new design Multiple DAC's are a fundamental part of Direct Interpolation. Many listening sessions confirm that a well designed multiple DAC system can sound significantly better than a single one.

- Jitter, I solved this by designing a custom made differential interface for the SPDIF signal, so no coax and no TOSlink. My opinion is to avoid jitter at the source. (asynchronous reclocking just seems to add jitter instead of removing it caused by the coincidental D-flipflop trigger). The higher the reclock frequency, the lower the jitter.

- I/V conversion, I used the classical OP-AMP approach, an OPA627, 470 Ohm and 220pF for this (shame on mejavascript:smilie(':whazzat:')). Passive I/V conversion using 33 OHms already creates a voltage drop at 4mA full scale of 132mV. Philips datasheets indicate that more than 25mV already causes distortion. Of course I also tried a tube output stage, it sounded nicejavascript:smilie(':)') but it doesn't solve the distortion problem caused by the I/V resistor and/or inductors (inductors seem to produce addittional noise). It also requires capacitors or transformers in the signal path. Personally I don't like non-linear components in the signal pathjavascript:smilie(':whazzat:').

- Clock syncing, I used separate 1% polystyrene 470pF capacitors on all TDA1541 (the close tolerance is important). It is used for internal timing of the current source switching. If you want to sync them use a 100pF between both pin's 16 instead of a direct connection, the oscillators run in phase but oscillator output amplitude may vary.

- Decoupling, The decoupling capacitors are just used for that, decoupling, not sampling. However, leakage currents cause massive distortion. High quality polyesterfilm or polypropylene will do fine. Keep wiring as short as possible.

- Differential mode, I inverted the data signal on one DAC. The first dac outputs L and R, the second one L- and R-. So basically I used 4 I/V converters that feed into a differential amplifier. Interference is significantly reduced this way and the output is DC-coupled since the offset voltage of both TDA1541A's is cancelled out.

- Filtering, I experimented for over a year with the most exotic filter setups. I finally used 2 closely tuned Allen key 3rd order butterworth-isch filters followed by a 2nd order MBF butterworth-isch stage. Filter was optimized for both flatness and linear group delay that's why I stated butterworth-isch. The sound is about the best you can expect from a 8th order filter. By the way, there is a nice filter calculation program available from TI.

- Printed circuit board, Over the many years I designed hundred's of circuit boards (without autorouter), so this is a piece of cake. The twin DAC was placed on a single circuit board, but the new quad Direct Interpolation DAC will be a modularar design to achieve optimal results and future updates.

Well as you can see I can get quite enthousiastic about electronics.....Well that was it for now, I am pleased to join diyAudio. If you want more information about my D.I.Y. projects, please feel free to ask, your comments, tips and reactions are very welcome.
 
Hi and Welcome.:wave: :wave:

It all sounds very nice but, as they say, a picture tells a thousand words! So, how about posting some pictures of your previous projects and, more importantly for this thread, some photos, schematics and more technical info of this project.

Andy:)
 
-ecdesigns- said:


- Differential mode, I inverted the data signal on one DAC. The first dac outputs L and R, the second one L- and R-. So basically I used 4 I/V converters that feed into a differential amplifier. Interference is significantly reduced this way and the output is DC-coupled since the offset voltage of both TDA1541A's is cancelled out.


Wellcome to the forum -ecdesigns-, enjoy!

My understanding is if you have L- and L+ from different dacs output will be out of balance somewhat unless you have perfectly match dacs. Another approache, that is allready cover in this forum, is spliting I2S signal in to left and right feeding two TDA1541A. That way you'll have L- and L+ output signal (or R- and R+) from the same dac, and the signal is perfectly in balance (hopefully). Current offset could be eliminated by offset nulling circuit, like 2m current source in front of I/V converters. For I2S spliting circuit see link.

http://www.diyaudio.com/forums/attachment.php?s=&postid=647561&stamp=1116801939
 
Twin DAC photo

Thank you for the replies,

I added a photo of the twin DAC, it is in use for over 3 years now, there are definitly no problems with DC component. I will post some more photo's and diagrams as soon as possible (I have to convert the diagrams first). The question about the differential mode: In differential mode the worst DC component I had was about 10mV. How is it done? First you have to INVERT the SERIAL data of the SECOND dac with an inverter (SN74HCT04 or so), clock and wordselect are the same. Then put an I/V converter on every DAC output (4X). Now the L- goes to the - input of the differential amplifier (classic set-up around a single OP-AMP). So +900mV with a digitally inverted sample. The L+ goes to the + input of the differential amplifier So 900mV with a normal sample. Out comes: +900mV (-) +900mV=0V DC. The Inverted sample is inverted again ( - input diff amp) and becomes positive, then it is added to the normal sample, so you get twice the amplitude. Do the same for the right channel. So inverting the SERIAL DATA on the second DAC makes this possible. The schematic diagram I will post later will explain this more clearly.
 
photo twin-DAC

Photo was still too large, I rescaled it again
 

Attachments

  • twindac.jpg
    twindac.jpg
    55.8 KB · Views: 68,194
photo Sonic Resonator

Here is a photo of the sonic resonator, a 360 degree radiator with real-time correction system for sub-bass, bass and mid. This system needs no damping materials. Mid-high and High are placed in an open resonance chamber with conical deflectors and polarizer discs for incredible realistic sound, it actually reconstructs the acoustics of the place the recording was made, with pinpoint accuracy. Chesky audio recordings sound magnificent on these sonic resonators. The photo shows an early prototype of about 4 years ago. Components are handcrafted form MDF and sprayed with metallic paint. The new version has solid aluminum and plastic parts and is sonically optimized to the max. It has a highly optimized passive filter using some very exotic parts like NON-inductive stranded copper wire resistors (took a long time and many listening sessions to fine-tune). You cannot believe how much effect a simple component as a resistor, capacitor or coil can have on sound reproduction. As I already said, hearing is believing. The most common reaction I get from High-End enthousiasts is: this sound cannot possibly come from those speakers....but then, it does. Second is, where is the subwoofer? ,those speakers cannot produce that bass or can they... sure they can.
 

Attachments

  • sonic resonator.jpg
    sonic resonator.jpg
    71.4 KB · Views: 61,015
  • Like
Reactions: 1 user
Photo Quad DAC module

NOS DAC's have one little problem: ultrasonic interference. The mirrored frequency spectrum above the sample frequency causes interferences and a drop in the high frequency audio range among other things. But still they sound so nice. First I designed the twin DAC wit an 8th order active filter, this was a real big improvement, but... that filter kept bugging me. Imagine how a NOS-DAC would sound without this interference, after many sleepless nights (I never rest before a problem is solved) I solved it in an elegant way. Here is a photograph of the Quad DAC module, part of the new NON oversampling DAC with Direct Interpolation (other circuit boards are being designed). The prototype is already operational (spaghetti breadboard setupjavascript:smilie(':D')). This NON oversampling DAC has 176.4 KHz virtual sample rate. Virtual? yes, because all DAC chip's still run on 44.1 KHz, but the output signal sports a 176.4 KHz sample rate. The resolution (without any filtering) is boosted to 18 bits. The output voltage is qadrupled. There are no phase shifts since there is no filter, it is true phase linear. Pulse responce is what you expect from a NON-oversampling DAC. Needless to say that dynamics are impressive using 4 DAC's. But how is this possible? by using Direct Interpolation. It uses timing tricks and analog wizzardyjavascript:smilie(';)'). I added a 2nd order bessel filter to avoid possible problems, but on my set it works perfectly without any form of filtering. This DAC sounds so clear without the ultrasonic interference and the absence of a brickwall filter, it makes the TDA1541A really shine as it has never done before. Is this DAC very complicated?, well actually NO. For each problem there is a complicated and a simple straigt forward approach. This is the simple straight forward one. I can post a picture of the breadboard prototype, but be warned, it is not a pretty sightjavascript:smilie(':)').
 

Attachments

  • quaddac.jpg
    quaddac.jpg
    63.2 KB · Views: 62,008
  • Like
Reactions: 1 user
Re: Photo Quad DAC module

-ecdesigns- said:
............ Here is a photograph of the Quad DAC module, part of the new NON oversampling DAC with Direct Interpolation (other circuit boards are being designed).
......... I can post a picture of the breadboard prototype, but be warned, it is not a pretty sight .


Please post a photo and schematic.

Andy
 
Hi ecdesigns,

This is great! I recently posted a thread asking if anyone has tried paralleling TDA1541 in a balanced configuration. As far as I know, you are the only one who is actually implementing this. Please let me know if you are planning to build a kit for sale. I would jump at the chance of buying it. How many 1541s can be practically paralleled in your design? Could it be piggy backed instead like the well known DDDAC1543 (www.dddac.de)? Good work!
 
breadboard prototype

Hi Andy,

Here is a photo of the new Quad DAC on the breadboard, below it is the High-End set I designed. On top is the input switchbox, in the centre is the 6 channel control amp, at the bottom is the fully symetrical rock-solid MOSFET cascode amplifier. I tried to post the schematics, but the file is to big, if I scale it down it's unreadable. I could email it and attach the original file (about 1,5Mb).
 

Attachments

  • breadboard.jpg
    breadboard.jpg
    32.1 KB · Views: 31,352
Direct Interpolation

Hi, Uchi Deshi

Thank you for your reply

Direct interpolation uses at least 2 DAC's, this creates 88.2KHz virtual sample rate and doubles the resolution. The Quad Direct Interpolation DAC creates 176.4 KHz virtual sample rate and results in 18 bit resolution. And yes sure you can use more DAC's in parallel, 8 DAC's will create 352KHz virtual sample rate and so on. How does it work? It involves slight timeshifting (delay) of both Data and Wordselect and connecting 2 pair of DAC outputs in parallel, then these two are fed into a differential amplifier after I/V conversion. So this system only needs 4 I/V converters. The easiest way of visualizing this system is comparing a 1 cylinder engine with say a 4 cylinder engine. The 4 cylinder engine runs much smoother and achieves much higher revs than a 1 cylinder engine would. Now the DAC's are the pistons, the crankshaft is the output stage and the ignition is the timeshifting. So all 4 DAC's receive their DATA in sequence (not simultaneously). So they output their sample on slightly different moments in time (one clockpuls after wordselect changes) and thereby creating addittional interpolated steps i.e. increasing the resolution. So basically it's a quite simple technique. Since an added sample always stays in the timeframe (64 clockpulses) of the original (large) sample, there is no phase shift. Schematic diagrams will follow.
 
SPDIF (white connector)

Hi analog sa,

The white connector you referred to is actually the analog output. I use a custom made fully differential SPDIF interface like in professional sound studio equipment. So no 75 OHm coax cable, certainly no TOSlink, actually I am using standard SVHS cable for this purpose. So I have a normal and an inverted SPDIF signal to cancel out interference that causes jitter. Obviously I had to modify my CD player (added schmitt trigger differential stage) so it would output this clean jitter-free differential signal to start with. Why a schmitt trigger? to get very short rise times as long risetimes can cause jitter. And jou must make sure the CS8412 get's a stable low jitter signal, otherwise its PLL keeps fluctuating and makes matters worse. So if you want to get rid of jitter, tackle the problem by the source and go differentialjavascript:smilie(';)').
 
Linear interpolation

Hi rfbrw,

Thanks for your reply,

You are right, the prototype uses a form of linear interpolation, but without the traditional form of digital filtering, no decimating, no brickwall filtering, and no phase shifts. So I called it Direct interpolation as interpolation takes place instantly without computations. The beauty of it is that you have a filterless NON oversampling DAC outputting 88.2KHz, 176.4KHz or even more. A pure clean signal. Please read my other posts were I explained the system.