Building the ultimate NOS DAC using TDA1541A

Hi ernesternest,

I'm using your MK4 I/V stage and have implemented your stepped rectifiers.

I made a mistake with the FET output stage.

Problem is that the output stage has very large bandwidth (theoretically up to FET Ft). This is a property of the grounded gate buffer. FET input capacitance (approx. 30pF for 2SK170) is placed in parallel with FET current buffer input impedance of around 12 ... 40 Ohms. The result is that the FET is always operated close to its maximum frequency and thus introduces distortion and causes instability. This can lead to audible distortion, and that's basically the flaw that was in all of the current buffer circuits I used.

When using slower (lower bandwidth) output circuits like OP-amps or discrete OP-amps, the problem also occurs.

Possible solution is placing a suitable capacitor (220pF ... 10nF) between DAC output and GND.

In practice the DAC outputs must always be band limited prior to feeding the signal to the output stage. So called I/V transformers provide required band limiting.

When using a (FET) current buffer together with the TDA1541A it's important to prevent the transistor or FET from switching off. This can happen when DAC output signal reaches zero (range 0 ... 4mA). In order to keep the transistor or FET conducting at all times, suitable bias resistor or CCS from DAC output to -15V must be used. After switching to MOSFET buffers with adjustable gate voltage, the value of the bias resistor could be increased as the DC offset can now always be set to zero, regardless of bias resistors. With 2SK170, the bias resistor also lowered the DC voltage on the DAC output by drawing extra current in order to meet JFET drain current when Vgs = 0.

With the MOSFET converter I can now use 150K bias resistor to -15V.

Regarding the stepped rectifiers you mentioned in your last post, the resistor values depend on the application. As far I've seen you've posted two different pictures so far. In one you used 10R and 47R, in an earlier scheme I think I saw 47R and 470R. Is it possible to give a rule of thumb for dimensioning these, lets say depending on the current drawn.

For low power applications (approx. 1 ... 10 watts) I use 68 Ohms and 680 Ohms. For the Circlotron amps (50 ... 100 watts) I use 10 Ohm and 39 Ohm. Exact values depend on smoothing capacitor ESR / impedance, and average load (bias) current.

This week I tested discrete series regulators, these were designed for large bandwidth, low noise and fast tracking. PCB size is small enough (11 x 28mm) to replace 78XX and 79XX regulators in existing designs. The regulators are based on 2 current mirrors (BCV61 / BCV62), 2 transistors, 5 resistors, and 2 LEDs (reference), so 11 parts in total. I designed both, a positive and negative regulator. Short circuit protection is provided by combination of constant current source driving the series transistor, and series transistor DC gain (around 40).

Sound quality of these discrete regulators is a lot better compared to IC regulators. It appears as if IC regulators color the sound, add grain and limit resolution.

Searching not only for the name but for function brought this:

That's the correct D flip flop.
 
Member
Joined 2007
Paid Member
Thanks John!

So understood flipflop is the one and now I know more about the stepped rectifier.


Hi ernesternest,

I made a mistake with the FET output stage....

Possible solution is placing a suitable capacitor (220pF ... 10nF) between DAC output and GND.

In practice the DAC outputs must always be band limited prior to feeding the signal to the output stage. So called I/V transformers provide required band limiting.

When using a (FET) current buffer together with the TDA1541A it's important to prevent the transistor or FET from switching off. This can happen when DAC output signal reaches zero (range 0 ... 4mA). In order to keep the transistor or FET conducting at all times, suitable bias resistor or CCS from DAC output to -15V must be used. After switching to MOSFET buffers with adjustable gate voltage, the value of the bias resistor could be increased as the DC offset can now always be set to zero, regardless of bias resistors. With 2SK170, the bias resistor also lowered the DC voltage on the DAC output by drawing extra current in order to meet JFET drain current when Vgs = 0.

With the MOSFET converter I can now use 150K bias resistor to -15V.

So it seems I had three options:
a) Put a 220pF and a resitor into the circuit
b) put a 220pF and a CCS into the cicuit
C) use Mosfets after your advise and plan.

In the end the third option might be the way to go. But I'd like to try the other ones as well. For b) How many mA the CCS should draw? 1-2, or up to 4? Then it could be another K170 as well? Or for a) which resistor value you would choose?

This week I tested discrete series regulators, these were designed for large bandwidth, low noise and fast tracking. PCB size is small enough (11 x 28mm) to replace 78XX and 79XX regulators in existing designs. The regulators are based on 2 current mirrors (BCV61 / BCV62), 2 transistors, 5 resistors, and 2 LEDs (reference), so 11 parts in total. I designed both, a positive and negative regulator. Short circuit protection is provided by combination of constant current source driving the series transistor, and series transistor DC gain (around 40).

Sound quality of these discrete regulators is a lot better compared to IC regulators. It appears as if IC regulators color the sound, add grain and limit resolution.

Hmm, sounds very promising. Have been looking around for some nice desing for a long time. Currently I'm working with Salas shunts and Super Teddys, but they are both not really suitable for just replacing ICs in existing designs.

Cheers Ernst
 
Hello John

Finding good quality Jfet are difficult here arround.

So I've mods an existing bjt I/V amp for a nos TDA1541A, there is an input and output 20khz low pass filters.

What do you think of my circuit ?

Thank

Bye

Gaetan
 

Attachments

  • IV amp-BJT FOR TDA1541A.gif
    IV amp-BJT FOR TDA1541A.gif
    65.3 KB · Views: 1,588
Hi gaetan888,

Finding good quality Jfet are difficult here arround.

I attached a quick screen shot of analogue signal path schematics of my set, no JFETs required here. I only have 5 power MOSFETs in the signal path between TDA1541A and speaker.

I/V converter is built around T1 that's operating in grounded gate mode. Many different power MOSFETs can be used here. P1 taps suitable DC offset voltage from passive I/V resistor R1, blocks the audio signal using C1, and feeds it to the gate of T1. The "interference" that passes C1 only introduces ever so slight negative feedback as the interference is the audio signal itself.

DC voltage at the DAC can be trimmed down to zero volts with multi turn trimmer P1. C2 band limits the TDA1541A output prior to feeding the power MOSFET current buffer. Bias resistor R2 prevents T1 from going out of conduction by drawing 100uA bias current. R2 value is high enough (150K) to minimize -15V rail noise injection. C3 blocks DC and is the only coupling cap in the signal path. One coupling cap is always required (safety) in case the I2S stream stops for some reason and the DAC chip latches the last received data.

This converter has approx. 14 Ohm input impedance, this results in approx. 56mVpp at the DAC output. 50mVpp would be ideal (+/-25mV compliance) but it comes very close. This converter feeds all bit currents into +5V DAC power supply for best performance.


Note that volume pot P2 is connected in parallel with the passive I/V resistor R1. GND of both converter and power amp must not be connected. This removes the DAC +5V power supply from the direct signal path. So I am basically "measuring" the ac voltage across R1 only.


The thingy on the right is my latest Circlotron MOSFET power amp. PowerFETs T1 and T2 form a differential input stage (gain = 15). P1 sets the bias current (and voltage drop across R5 and R6). This in turn sets the bias current in the Circlotron bridge output stage built around Power MOSFETs T4 and T5. P3 sets the DC-offset voltage on the speaker output (note that this power amp is fully DC-coupled!). C4 and R4 fixes the gain difference between T2 (grounded source, high gain) and T3 (grounded gate, lower gain). R7 and R8 provide reference for driving powerFETs T4 and T5.

The Circlotron bridge output stage doubles the output signal across the speaker (measured total gain equals 30x). With 0.7V rms DAC output this results in max. 21V rms across the speaker. This enables max. output power of approx. 55 watts rms in 8 Ohms or 110 watts in 4 Ohm.

Because of the use of power MOSFETs exclusively (substrate mounted on heatsink), thermal memory effects are minimized. The few components in the signal path ensure both, very high resolution and transparency.


So I've mods an existing bjt I/V amp for a nos TDA1541A, there is an input and output 20khz low pass filters.
What do you think of my circuit ?

BJT version of the ZEN I/V converter with added filters?

Its a clever design but has drawbacks. Filters are debatable, I try to avoid them whenever possible (phase shifts).

The TDA1541A operates optimally when all bit currents flow back into +5V TDA1541A power supply. This is not the case here.

You need 2 very high quality wire wound passive I/V resistors. These resistors make or break sound quality as they perform actual I/V conversion.

TDA1541A only sinks current and sinks 2mA bias current, this might affect circuit performance.

Both C1 and C2 are over dimensioned, lower value film caps can be used instead.

R12 injects significant power supply noise into the DAC outputs.
 

Attachments

  • ivcir.jpg
    ivcir.jpg
    85.1 KB · Views: 1,539
Last edited:
Member
Joined 2007
Paid Member
When using a (FET) current buffer together with the TDA1541A it's important to prevent the transistor or FET from switching off. This can happen when DAC output signal reaches zero (range 0 ... 4mA). In order to keep the transistor or FET conducting at all times, suitable bias resistor or CCS from DAC output to -15V must be used. After switching to MOSFET buffers with adjustable gate voltage, the value of the bias resistor could be increased as the DC offset can now always be set to zero, regardless of bias resistors. With 2SK170, the bias resistor also lowered the DC voltage on the DAC output by drawing extra current in order to meet JFET drain current when Vgs = 0.

With the MOSFET converter I can now use 150K bias resistor to -15V.

John, read again this post. So that means, the 5.6 K Resistor is not suitable to keep the JFet buffer working even the DAC current ist 0?

Regards Ernst
 
Hi gaetan8888,

I only have BS170R mosfets, is it ok ?

I've redraw your circuit so it be connect to the TDA1541A, to be sure I've understand it correctly.

I just tested BS170 in the converter and it works. Your redrawn circuit is ok.

I haven't re-checked this but the bias resistor (and resulting current) also affects converter input impedance. With BS170 output compliance of 50mVpp is fully met when using 5K bias resistor.

In practice it's a trade-off between meeting output compliance (low bias resistor value) and minimizing noise injection from the -15V rail (high bias resistor value). You can experiment with bias resistor values between 5K and 100K and settle for the value that offers best perceived sound quality.
 
Hi ernesternest,

John, read again this post. So that means, the 5.6 K Resistor is not suitable to keep the JFet buffer working even the DAC current ist 0?

When I used the JFET converter (2SK170), the JFET drain current was approx. 5mA (Vgs = 0). Since the DAC only draws 2mA bias current (current sink) I was missing 3mA. Failing to correct this will result in positive DC voltage on the DAC output (output compliance). The DAC output will be at 0V DC when the FET drain current specified at Vgs = 0, is exactly met.

I had no clean negative bias voltage available (power supply pollution) to pinch-off the JFET reducing the drain current to required 2mA.

So I used a 3mA bias current instead. The bias resistor also prevented current interruption through the JFET when DAC output current equals zero (DAC output is basically disconnected / floating in this case).

With the MOSFET converter I have a trimmer for adjusting source voltage to zero by varying Vgs. So the bias resistor is no longer required for trimming DC voltage at the DAC output to zero. It is however still required to prevent the MOSFET from going out of conduction (similar as with the JFET). But bias current can now be lowered depending on required converter input impedance.
 
Hi gaetan8888,

I just tested BS170 in the converter and it works. Your redrawn circuit is ok.

I haven't re-checked this but the bias resistor (and resulting current) also affects converter input impedance. With BS170 output compliance of 50mVpp is fully met when using 5K bias resistor.

In practice it's a trade-off between meeting output compliance (low bias resistor value) and minimizing noise injection from the -15V rail (high bias resistor value). You can experiment with bias resistor values between 5K and 100K and settle for the value that offers best perceived sound quality.

Hello John

Missing few parts to built it for now, but I've simulated it, using a 2ma sine current source to replace the TDA1541A

It's not as the real TDA1541A since that using more sine current source do overload it.

Here is the image and the Ltspice asc file into a text file.

Thank

Bye

Gaetan
 

Attachments

  • D1-IVamp-BS170-2b.txt
    2.9 KB · Views: 127
  • Capture.jpg
    Capture.jpg
    97.9 KB · Views: 1,390
Last edited:
John,
I am interested by your Circlotron deign. I though, I have read that you biased this amplifier in class A operation. Somewhere you said 55W output.
Here are my questions.
Examining the power supply section in your post 3467, I presume the 25va transformers are for a lower output amplifier?
The class A operation interest me, so dissipating 100W of heat with 2 devices will make the temperature rise to unsafe limits. Would it be possible to parralell a few 2SKs without reducing the performance of this amplifier?
What would be a maximum of devices before degrading the preformance?
Of course it will require sufficient heatsinking and adequate supply, but do you have any suggestions, or changes for such a project?
Thank you for sharing your experience with us.
 
Member
Joined 2001
Paid Member
Jean-Charles- back in post #3467 John mentions running 150mA through the outputs.

I'm also intrigued by this Circlotron (and EUVL's), and watching closely. I would like to try a circuit like this with some SemiSouth R100's in class A ("Sic Circlotron"?), driving the top of a bi-amped setup, ~300 hz on up. The wirewound pot is interesting also, comments of others? inductance too high.. rolls of treble... but their example is always a 100k pot.... 10k or 20k may be just fine? what about a 3 turn vs. 5 turn vs. 10, does it make any difference?

Seems like a balanced source is really what you would want driving a circuit like this. And class A operation. But, what do I know. I really need to get off my duff and build something, and snap out of the analysis paralysis.
 
Hi Jean Charles,

I am interested by your Circlotron deign. I though, I have read that you biased this amplifier in class A operation. Somewhere you said 55W output.

Circlotrons can work in both class A or AB. I run it in class AB (200mA bias).

Output stage runs on 2 separate 25V / 25VA power supplies, this consumes 10 watts during idling (5 watts for each FET). This way I get around 25 watts rms in 8 Ohms. Differential input / driver stage runs on 18mA bias (class A) and offers 15x gain. The Circlotron bridge output stage doubles it to 30x.

Max output power with 0.7V rms from the TDA1541A-MK4 module equals 0.7 * 30 = 21V rms. This results in max. 55 watts in 8 Ohms with given 15x input / driver stage gain.

Examining the power supply section in your post 3467, I presume the 25va transformers are for a lower output amplifier?

Yes, approx. 25 watts in 8 Ohms.

The class A operation interest me, so dissipating 100W of heat with 2 devices will make the temperature rise to unsafe limits.

You could try MOSFETs similar to these:

STMICROELECTRONICS|STE53NC50|MOSFET, N, SOT-227B | Farnell Nederland

Temperature compensation would also be required.

What would be a maximum of devices before degrading the preformance?

One MOSFET for each bridge half. I tested multiple matched MOSFETs in parallel, but this resulted in degrading (loss of resolution and transparency). Note that this design has only 4 power MOSFETs in total and has no global feedback to cover-up flaws.

Of course it will require sufficient heatsinking and adequate supply, but do you have any suggestions, or changes for such a project?

Higher power consumption will also significantly increase power supply noise (transformer EM fields and rectifier switching noise). Stepped rectifiers would be a must, and the transformers need to be placed further apart to avoid coupling between the EM fields of both transformers powering the Circlotron output. In order to minimize magnetic coupling it would be best to use toroidal transformers that offer low external magnetic field.

Note that with a circlotron power amp the speaker (output) sits between both power supplies. Any (capacitive) coupling between both power supplies would then occur in parallel with the speaker / output.
 
Hi William2001,

The wirewound pot is interesting also, comments of others? inductance too high.. rolls of treble... but their example is always a 100k pot.... 10k or 20k may be just fine? what about a 3 turn vs. 5 turn vs. 10, does it make any difference?

The inductance is rather small (uH range), too small to cause trebles roll-off. Both inductances are placed in series (input-wiper, wiper-GND) so impedance increase should have little effect too.

Only problem is that they are linear. With a 1 turn pot it's difficult to set lowest volume. Multi turn pots offer better control at lowest volume settings. Inductance of multiturn versions is comparable with single turn versions.

I'm also intrigued by this Circlotron (and EUVL's), and watching closely. I would like to try a circuit like this with some SemiSouth R100's in class A ("Sic Circlotron"?), driving the top of a bi-amped setup, ~300 hz on up.

No problem, you might need to change value of R3 (250 Ohms) in the latest schematic to achieve correct bias setting range. Temperature compensation (bias current) would also be required to prevent thermal runaway.
 
Discrete mini regulators

I attached some pictures of the completed discrete regulators that I plan to use in both, TDA1541A-MK4 and Circlotron.

I made a positive (78XX) and negative (79XX) version. The PCB measures approx. 11 x 28mm. Voltage is indicated by a color on top of the PCB.

Both voltage and max. output current (short circuit protection) are set by resistors. There are 2 sets of pads for voltage setting resistors for more accurate voltage trimming.

The regulators are pin-compatible with 78XX and 79XX regulators and can be mounted on a heatsink for higher output currents.

Low noise references (zener / LED) are used for CCS and error amp. These produce around 500nV noise, much lower compared to bandgap or XFET references.

Capacitors could be left out (low noise references and low impedance voltage divider) for achieving lowest possible sound coloration.


I also placed an extra filter between -15V rail and 150K bias resistors in the TDA1541A-MK4 module for better performance. The filter consists of 22K resistor in series with 10mH choke and a 470uF cap.
 

Attachments

  • minireg.jpg
    minireg.jpg
    101.3 KB · Views: 1,206
TDA1541 in Simultaneous mode

Hello to everyone..

I'm writing here because it seems the guru's place about the TDA1541..!!:)

I would like to run it in Simultaneous mode in a Philips 304..

Is it possible to use the DEM from ECDesign for this configuration..?

ECDesign...do you hava any comment or suggestion for me?

Thanks to everyone will helo me...and regards from Italy..

Maurizio.
 

Attachments

  • screen-capture.jpg
    screen-capture.jpg
    145.3 KB · Views: 1,087