Building the ultimate NOS DAC using TDA1541A

One can set up theories pro and contra about anything in digital sampling . For example get samples of a 22.05 kHz sine wave. If you are unlucky, you might get samples at zero crossings, so the sampled and reproduced signal will become zero. Shift the analog signal by 90 degrees with reference to the sampling clock, and you'll get full amplitude samples. Play back this digital stream on a NOS DAC and you'll get square wave. Use a brickwall filter @ say 25kHz and you'll get perfect sine wave. You get exactly the same results when the original analog signal were 22.05 kHz square instead of sine (although in this case you can't get samples at zero crossings since such do not exist).

I just wanted to illustrate with the above reasoning, that theories and single-signal measurements could be misleading. It is listening to real music that will tell more about the truth. Unfortunately the measurement instrument in this case is the human ear, which is uncapable of giving numerical results, and is influenced by many psychoacoustic factors.

If you want to get objective measurement results, you have to use a test signal close to real music program material. Such test signal IMHO is pink noise, combined with an FFT analyzer.
 
One can set up theories pro and contra about anything in digital sampling .
For example get samples of a 22.05 kHz sine wave. If you are unlucky, you might get samples at zero crossings, so the sampled and reproduced signal will become zero.


The frequency response must not be corrected because it might be flawed by sampling anyway ?
If that is your logic, then you could use opamps everywhere because opamps might have been used on the recording side.
Also why care about the quality of the speakers, you don't know how good the microphone was.
And hey - how jittery was the sampling clock ? So a crystal with a logic gate inside your cdp should perhaps do it for you.

Recordings are of the quality: "as is"
Our task is to convert the "as is" data into analog signals as accurate as possible.
That includes both a flat frequency response and a waveform that does not look like a staircase.
 
Bernhard, don't suggest we disagree :). Again the good old NOS or no NOS question. NOS for me sounds better (after all this is why I follow this thread with most interest) and to be honest I haven't tried a sin(x)/x filter yet. For me it sounds good enough without, and I am not convinced it is necessary at all. Only a properly done measurement will convince me. Properly I mean not a swept sine but some complex signal I referred to before.
 
filters again

Hi Oshifis, I agree with you.

Hi Bernhard,

I've done a lot of tests with this IC (1541). Op amps, no op amps, tubes, SS, filter, no filter, reclock, no reclock......., etc. "My personal" conclusion is: Even if the signal looks like a "staircase", our ears are not able to discern such low levels of "stairs". At this moment, I have an old Philips CD player with: my own superclock design+ECD output (I/V concept based)+ECD signal treatment, it is not the final version, but I've put it against a very high level equipment (not going to say names), and I prefer always the naturality of the TDA with it's response (seen from the osciloscope). The sound is extremely detailed, never fatiguing, "crystal clear", the bass is "incredible detailed" (tested with Pirates of the Caribbean OST, an extremely difficult one), the voices are naturals and sweets, the percussion, simply astounding, hiperdynamic, and, I can assure you that I have very good ear (I've played music in a group for years). I believe here we are treating to obtain the best from an old DAC, in fact, the most incredible and fine DAC I've heard. We don't want to argue against this technology, or the modern one, just have fun, learn and build the best dac we can.

With a total of 4mA/65535 we have a tiny 61pA to deal with. We should first exploit the full potential of the 16 bits, and then, build something better, but I think it is very difficult to achieve. The technology of 24 bits is completely unattainable, because if it costs us to obtain reasonable levels of noise with 16 bits, imagine with 24 bit. This patch is not really an improvement.

I expect this is the last time I have to argue this question. If you have a better solution, then send us an schematic or something simmilar that actually exceeds the TDA1541 technology, from the point of view of our ears. I'm open minded to all the new things, but only if it's really better.


Best regards to all.
 
Hi Oshifis, I agree with you.

Hi Bernhard,

I've done a lot of tests with this IC (1541). Op amps, no op amps, tubes, SS, filter, no filter, reclock, no reclock......., etc. "My personal" conclusion is: Even if the signal looks like a "staircase", our ears are not able to discern such low levels of "stairs". At this moment, I have an old Philips CD player with: my own superclock design+ECD output (I/V concept based)+ECD signal treatment, it is not the final version, but I've put it against a very high level equipment (not going to say names), and I prefer always the naturality of the TDA with it's response (seen from the osciloscope). The sound is extremely detailed, never fatiguing, "crystal clear", the bass is "incredible detailed" (tested with Pirates of the Caribbean OST, an extremely difficult one), the voices are naturals and sweets, the percussion, simply astounding, hiperdynamic, and, I can assure you that I have very good ear (I've played music in a group for years). I believe here we are treating to obtain the best from an old DAC, in fact, the most incredible and fine DAC I've heard. We don't want to argue against this technology, or the modern one, just have fun, learn and build the best dac we can.

Best regards to all.

Hi

Is your NOS TDA1541 dac are filtered or not filtered ?

Did you try the DAC using four TDA1541 with analog extrapolation project by Ecdesigns ?

Thanx

Paul
 
NOS vs OS

Unfiltered NOS offers highest accuracy in the time domain and poorest in the frequency domain.

Filtered NOS offers reasonable accuracy in both time and frequency domain.

OS offers highest accuracy in the frequency domain and poorest in the time domain.

Linear interpolation (multiple DAC chips in parallel fed by delayed I2S signals) works excellent at lower frequencies, but distorts higher frequencies causing early trebles roll-off among other things. It can work very well in combination with a digital brickwall filter like used by Cambridge Audio in their 4 x TDA1541A CD player.

The question would be what's most important for realistic sound reproduction, highest precision in the time domain or highest precision in the frequency domain. Based on many years of experimenting and listening to equipment designed by other developers it seems that accuracy in the time domain is much more important for achieving most realistic sound reproduction than we realize.

This is the reason why I choose unfiltered NOS. Compromise I have to make is reduced accuracy in the frequency domain as Bernhard pointed out.

The problem is, we can't have both, using 44.1 / 16 format, it's either one, the other or a combination of both. It looks like we have to make our own choices based on our personal preferences as there is no perfect way to go.

We also have to realize that sound quality is determined by audio component matching. Connecting some random equipment together is very likely to cause mismatches and resulting sound quality degradation. It is also very important to understand specific audio component properties in order to achieve optimal matching.

Example, unfiltered NOS won't work optimally with class-D or comparable switch-mode power amps as frequency spectra of both DAC and amplifier will inter-modulate.
 
NOS configuration

Hi Tessier, I hope you are doing fine,

I've tested the 4xDI with TDA1541, and ECD is right, it is not as good as 1 unit alone. My actual configuration is: CS8416+ECD signal treatment+my own superclock design+ECD IV concept output. This configuration is very very precise and detailed, never fatiguing. Like I've told before, it's incredible. It sounds very natural. I've tested it with a lot of CD discs. Try to test it with Karma, by Mars Lasar, and Pirates of the Caribbean or Gladiator, for example. You will see the hyper-dynamic, detailed, natural, musical, etc.

I've tested against a high level equipment, and I still prefer it. Imagine the final prototype well done, not a CD based one.

Best regards,
 
I am staying with my 4x 1541-S1 Linear Interpolation Dac...I try 1 to 4 DAC chips. My vintage Marantz system with Ohm Fs seems to work very nicely with the 4x chip NOS. Did my best to reclock the DAC.

Time domain correctness.... Well the Ohm Fs are about as time domain correct as you will get.

jk
 
Hi jstang,

I have been experimenting some more with multiple TDA1541A chips in parallel lately (no interpolation). Linear interpolation results in trebles roll-off and increased distortion at higher frequencies when compared to plain NOS. As for the resulting sound, I demonstrated my DI DACs on the HiFi 2007 show in Belgium and for a number of Dutch audio clubs. Only negative remark I got is that the DACs revealed too much detail, none of the listeners complained about trebles roll-off nor distortion.

With the SD-transport as source (< 2.5ps rms at DAC I2S inputs) paralleling multiple TDA1541A chips leads to less refined sound. Same happens when paralleling TDA1543 DAC chips. It is also obvious that when paralleling multiple stereo DAC chips the channel separation gradually degrades as L / R channel crosstalk between multiple chips increases. Similar, the required logic circuit on each chip will produce certain amount of interference, paralleling multiple DAC chips will result in summing logic circuit noise levels. Other problem (that can be corrected to some extent) are increased tolerances between L and R channels.

The TDA1541A introduces an extra problem in the form of the extremely critical DEM circuit that introduces even more problems when paralleling multiple chips (synchronization). Unfortunately this DEM circuit has big impact on bit errors, so it is very important to get this circuit functioning perfectly before even thinking of optimizing other circuits.

So far I have tested all DEM circuits published on the net, including Philips data sheet app using over 200 different capacitance values (trimmers). I also explored the DEM "locking" that occurs when the DEM oscillator frequency is (almost) an exact multiple of fs). None of these solutions offered desired results. I now use balanced DC-coupled zener diode, diode, resistor circuit that is driven by an ultra fast CMOS divider (705.6 KHz) running on 2.5V. This way I can achieve extreme low DEM clock jitter levels. and minimum noise injection at the DEM pins.

TDA154x chips have on-chip main current sources, one for each channel. When paralleling multiple DAC chips one cannot guarantee that power supplies have exactly the same DC voltage and interference (noise) levels and spectra. This could lead to increased (dynamic) bit errors (bit currents vary during the time the sample should remain fully stable).

Similar, chip temperature of paralleled chips will vary, this would lead to varying performance (dynamic matching).


The only possibility I could think of that wouldn't have above issues is using TDA1541A chips in dual mono mode.

Now channel separation even increases as each chip processes only one channel. The logic circuit on the chip produces similar interference as when running in stereo mode. The tolerances between L and R channels are reduced as each chip processes one channel only. Bonus is also the doubled full-scale output current that allows for using lower value I/V resistors (lower output impedance). DEM clocks don't need to be synchronized as both DAC circuits for one channel now sit on the same chip. Paralleled DAC circuit temperatures would closely match / track. Power supply DC voltage, interference levels and spectra would closely match too.

I tested this dual-mono mode recently, using a simple I2S encoder that converts L/R to L/L for chip #1 and R/R for chip #2. Both chips receive same BCK and WS signals, only the DATA signal for each chip differs. I didn't use balanced mode so possible "issues" with inverting DATA wouldn't occur.

The outputs of each chip were buffered (2 JFET current buffers + bias circuit / chip) and the drains of these buffers were connected in parallel and were used to drive the 250R I/V resistor providing 2Vpp (0.707V rms).

Ech chip has own set of decoupling caps, and I used latest 705.6 KHz low impedance DEM synchronizers (<2.5ps rms jitter).


Direct comparison with the latest TDA1541A-MK3 revealed no day and night differences. Very subtle differences were only audible by direct comparison and with specific high quality recordings. With poor quality recordings there was no detectable difference. This is also the "area" where psycho acoustics come in to play. Then one "hears" a difference when one knows exactly what DAC is playing. This can be solved by letting one person operate the switch and the other listens. The listener should not know what DAC is selected (indicators should not be visible).


Issues with the Dual mono mode are:

I2S source compatibility (I2S encoder must be adapted to support 32, 48, and 64 bits / frame).
Doubled bit and WS clock loads.
Doubled DEM synchronizer load.
Extra interference introduced by the I2S encoder circuit.
Increased cost / complexity.
Increased power consumption / power dissipation.

So one has to carefully evaluate pro's and con's of single vs dual DAC chip design.

I have made the decision to go for the single DAC chip approach in the form of the latest TDA1541A-MK3 DAC module.
 
I don't have test equipment to prove what I am about to say and I do agree with your statement below....

The one thing I would add... is that the glory of a good TT and cartridge is not its ability to reproduce a 20Khz square wave.... They can not.

So the resultant loss of what I would call extreme "hi freq" is meaningless to my ears too...as you noted below in the testing with some audio clubs members.

I think the DI DAC design comes closer to the Analog limitations of a TT and Cartridge without introducing time domain errors & phase shifts of filters.

The DI provides the same chip time delay smoothing for 20 Hz or 15K hz.... Which in the end is what I think sounds so sweet to my ears with the Ohm F speakers....

F's are a single VC with no crossover and are Time Coherent by design. They introduce no phase or time delay.

Which is why I am with you 100% on the Time Domain being very important... more so than extreme Hi Freq errors...... I just think its more important to be accurate in the sweet spot of the Ear's frequency range.

my 2 cents....


johnk


Hi jstang,

I have been experimenting some more with multiple TDA1541A chips in parallel lately (no interpolation). Linear interpolation results in trebles roll-off and increased distortion at higher frequencies when compared to plain NOS. As for the resulting sound, I demonstrated my DI DACs on the HiFi 2007 show in Belgium and for a number of Dutch audio clubs. Only negative remark I got is that the DACs revealed too much detail, none of the listeners complained about trebles roll-off nor distortion.
 
The one thing I would add... is that the glory of a good TT and cartridge is not its ability to reproduce a 20Khz square wave.... They can not.

So the resultant loss of what I would call extreme "hi freq" is meaningless to my ears too...as you noted below in the testing with some audio clubs members.

I think the DI DAC design comes closer to the Analog limitations of a TT and Cartridge without introducing time domain errors & phase shifts of filters.

Very well said, Sir.
Simpler is better.

So John, if you are not using your "scrambler/interpolator" then my DAC is the only one in use in the planet?? :cool:

Your honeycomb resistor idea is GREAT! Now I am listening to a passive I/V M1 DAC, with your low noise Vrefs: yammy, yammy...
 
Hi maxlorenz,

So John, if you are not using your "scrambler/interpolator" then my DAC is the only one in use in the planet??

Other diyAudio members have built and are using the DI DACs too.

The D-I system has the advantage that most audio equipment handles the output signal of these DACs without problems (smoother signal due to interpolation). The very important midrange is reproduced with lowest possible distortions in the time domain. The downside is trebles distortion / roll-off.
 
Hi Sandor,

I was just going to build the DEM clock as per MK2 version. Now I read that a new version is available.
Could you please post the schematic of the latest 705.6 KHz low impedance DEM synchronizer?

I attached latest TDA1541A circuit schematics:

- DEM synchronizer circuit
- I2S attenuators
- I/V converter

D flip-flops must be able to run on approx. 2.5V power supply. I use fastest CMOS logic I could find (800ps propagation delay @ 15pF load and 2.5V power supply) in order to minimize additive jitter.

Zener diodes are 400mW.

I2S attenuator has thermal tracking (diodes) and output levels are 0.6V (L) and 1.9V (H). They reduce ground-bounce in the on-chip I2S input circuit and as a result of this minimize I2S crosstalk and jitter.

I/V converter uses a plain (unselected) 2SK170-BL JFET and a 10K bias resistor. Actual I/V conversion is fully passive as the JFET merely passes the DAC output current as is. Gate current is much lower than LSB current.

I use +5V as reference instead of GND. This excludes +5V power supply from the direct signal path as I measure directly across the I/V resistor.

This converter provides simplest and cleanest output signal as the actual I/V converter consists of a single 500R resistor. All other options require active amplification and / or transformers that will introduce additional distortion.

Output coupling cap needs to be added to block DC. The hybrid coupling cap that performs closest to DC coupling was composed by paralleling 2.2uF/250V Monacor MKT, 3.3uF/250V Monacor MKT and a 40pF Teflon trimmer (set to max. value).
 

Attachments

  • tdacir1.jpg
    tdacir1.jpg
    150 KB · Views: 2,063
Hi gaetan8888

Have you compared trebles distortion between the DI4 and the DI8 ?

Distortion / trebles roll-off increases with increased interpolation factor (I tested up to 32x linear interpolation). I attached a collection of oscillograms that show interpolation at lower frequency (top left), higher frequency (top right), the effect of interpolation on the step shaped signal (bottom left) and impulse response (bottom right).

The problem is that with increased frequency, lower amount of samples are available for reconstruction (top right). Linear interpolation then tends to create triangle wave shapes as can be seen on the oscillogram. These have less energy (form factor) compared to a sine or square wave. This reduced energy level at the speaker results in reduced cone movement at these frequencies as less power is available. This results in trebles roll-off.

When using higher native sample rates (192 Khz for example), the D-I system works much better as there are more samples available to restore highest frequencies. Reflected mirror images are moved further away from the audio spectrum.
 

Attachments

  • oscdac1.jpg
    oscdac1.jpg
    83.2 KB · Views: 1,902
I am using 4x 1541-s1 in a DI dac with a CD-Pro transport. And simply love what it has done for the detailing of the mid range.

Again the speakers and gear I use are vintage. The speakers are real Walsh design Ohms Fs.... one VC and no crossover.

The treble roll off to my ears, on my system...was actually a cleaning up of the treble.

The OHM F's are one of the very few speakers that can reproduce a square wave. Perhaps that is the reason why I like the DI so much.

The treble lost its CD "square wave" harshness/edginess in my system. It did not seem to roll off, but it became cleaner and more defined to me with less edginess.

In back to back testing with my TT ( with a master vinyl recording ) and good CD..... The results very very close.....

The CD-Pro/DI-4 wins in some areas and the TT wins in others. Some 1960's recordings remastered on CD sound simply marvelous ! New stuff sounds great too....

But I grew-up with Vinyl and a good TT in the late 70's and 80's.... The CD-Pro/DI-4 ( with lots of reclocking ) has my great appreciation... I am not going to say I prefer it over vinyl.... nor would I say I prefer vinyl over it....

But the DI design puts CDs as close to vinyl as I have heard......with some pluses in the usual CD strengths area.

SD drive is next.... :D

johnk





Hi maxlorenz,



Other diyAudio members have built and are using the DI DACs too.

The D-I system has the advantage that most audio equipment handles the output signal of these DACs without problems (smoother signal due to interpolation). The very important midrange is reproduced with lowest possible distortions in the time domain. The downside is trebles distortion / roll-off.
 
Last edited:
Member
Joined 2009
Paid Member
I/V converter uses a plain (unselected) 2SK170-BL JFET and a 10K bias resistor. Actual I/V conversion is fully passive as the JFET merely passes the DAC output current as is. Gate current is much lower than LSB current.

I use +5V as reference instead of GND. This excludes +5V power supply from the direct signal path as I measure directly across the I/V resistor.

This converter provides simplest and cleanest output signal as the actual I/V converter consists of a single 500R resistor. All other options require active amplification and / or transformers that will introduce additional distortion.


Hi John,

i think the value of 500R must only be for usage with one TDA1541A.
If i have two of them, must i use 250R?

Best,
Oliver
 
Hi EC,

Other diyAudio members have built and are using the DI DACs too.

I know. I was asking specifically about the "scrambler/interpolator" one. Are there other builders?
I like it very much, though it has more chips on the signal path, but it has balanced option and is DC coupled. ;)
I am trying to find the time to make the last upgrades to it: better PSs and better Vrefs.
 
Hi dvb-projekt,

i think the value of 500R must only be for usage with one TDA1541A.
If i have two of them, must i use 250R?

With 2 x TDA1541A you simply duplicate the FET current buffer (one 2SK170-BL JFET and one 10K bias resistor for each DAC) and connect both drains of the JFETs to the I/V resistor. The full scale current now doubles to 8mA, so R = 2 / 0.008 = 250R.