Current-to-Voltage conversion is probably one of the most important circuits of any Current Output DAC. After all, this is where your analog audio signal is born (or I should probably say re-born) and any imperfections introduced here are impossible to eliminate down the chain.
To better understand the engineering challenges for this type of the circuit, let’s set couple things straight right ahead:
Now let’s look how this current is generated in TDA154x DAC’s. Above is only conceptual schematics, I did more in-depth explanation here. What becomes very obvious right away is how demanding current supply capabilities for +5V supply is! All bit switches are switching at the sample rate but the switching times are in nano second range! And they all happen at the same conversion instance – simultaneously.
But what if we could cheat and return all bit currents to +5V? Now all bit switches are always connected to the same power supply and sum-current are held constant. Neat! Except that in reality we must also keep output pin at GND level. Furthermore, +5V has now become our reference for output voltage. Otherwise all the switching noise from +5V will add to an output signal. So it’s not that simple as drawn above. Summarizing what was said we can state that:
That is unless this power supply is capable of maintaining exceptionally low impedance over audio range when presented with HF out-of-band signals. So choose your poison. If you use +5V as reference, than any low noise of-the-shelf voltage regulator will suffice and will sound excellent (with proper decoupling). I would advice this “almost free” mod for all of you, who are still looking how on how to squeeze more performance out of your CD player or DAC. Otherwise some kinda super-regulator are mandatory here.
And last but not the least, it would be nice to center output voltage at 0V to minimize distortion, so:
Just throw an trans-impedance op-amp at the output and call it a day! Well, if you are reading this, then I assume you have already tried that and it didn’t go well. Let’s see how we can improve on that.
This is what I would consider minimum circuit for meeting all the above requirements. And it’s really minimalistic, just one active device! Here a JFET is working in grounded grid mode as level shifter, so we can have 0V at analog output pin and still connect I/V resistor to +5V supply. JFET can be any high Gm low noise part (2SK170, BF862, 2SK2394, NSVJ3910). They all have enough trans-conductance to keep Iout voltage below ~150mVpp (35Ω resistor equivalent).
Distortion is quite high with 2nd harmonic going above -50dB (sorry, didn’t saved this measurement) but otherwise this simple circuit sounds excellent and is mainly limited by the quality of coupling cap. Another limitation is output voltage swing of only about 2Vpp. Going higher than that, increases distortion even more as JFET is operated further out of linear region.
And here is the classics of a genre. This is probably how 80% of TDA DAC’s are operated. At least in the DIY community that is. I/V resistor followed by a Tube gain stage. If using resistor or CCS load as shown above, output impedance is unexceptionably high (tube Re) so some sort output buffer must be used.
Distortion profile depends on what load is used for the tube. Tube linearity is also important and with triode strapped D3A and CCS load, 2nd harmonic is only at -80dB. Now output voltage is not a limitation and standard 2Vrms output can be achieved with only 25Ω I/V resistor. With 80Ω and CCS load, we can achieve 20Vpp output with really low THD. So it’s even possible to drive speakers directly (mosfet follower anyone?).
Unfortunately previous I/V stage doesn’t met the requirement for bit current return path. Again, how much is this a problem depends on your +5V supply quality. Above is the solution for bit current return to +5V. It’s not visually intuitive when drawn like that, so I added markers with DC current loops. It’s basically our first Single Grounded-Gate JFET I/V stage, but with a much lower I/V resistor value, followed by a tube gain stage.
Linearity can be improved by using CCS for Tube loading and Bias current to -15V. As can be seen from FFT plot above, using +5V as output reference has no influence on noise floor and doesn’t introduce additional spectral “garbage”. Otherwise distortions are almost the same as with “standart” 25Ω Tube I/V circuit. Slight increase in 2nd harmonic is due to a larger equivalent I/V resistance (~29Ω) of 2SK2394-6 JFET at DAC output. But sonically this circuit is a day and night difference in comparison to standard GND referenced I/V circuit. I’m always shocked how much things improve from this mod in any CD player with standard 78l05 voltage regs and poor decoupling. Don’t take my word for it, try it your self!
Up until now we were using our tube as a voltage gain device. And it does wonderful job at it for any audio signal. Unfortunately we ignored our first thesis of this article: “Signal (current) coming out of the DAC “analog out” pin is not analog at all”. I cannot stress enough how important it is. Now anyone with any background in ham-radio or any other HF related stuff knows how tubes (or any other gain device) should be used when dealing with high bandwidth signals. Grounded grid!
This way we say goodbye to pesky miller capacitance and now our tube works as a current conveyer for 1k5Ω I/V resistor. From the schematics above we see that our tube and it’s power supply are in the main current loop. Decoupling of this 240V power supply presents a challenge as it now must be able to return fast bit currents into the +5V supply. Good quality super-cap bypassed with silver mica does a wonderful job and costs equally astonishing as well. Unfortunately, only other option is to use super-reg’s for both 5V and 240V power supply’s.
One thing that catches the eye straight away is a much cleaner noise floor in comparison to gain-based I/V stages. This is partly because tube high impedance input (grid) is grounded and can’t pick up any noise. Also some inter-modulation with HF garbage mechanism could be at play here. Either-way it’s an astonishing overall improvement visually and sonically. Although I don’t attribute this sonic improvement to just cleaner spectrum. As always, there is much more at play here then is visible. If I had to guess, x10 bandwidth improvement over common-cathode gain stage has more influence here.
Another benefit of grounded-grid stage is less demanding tube selection. Pricey D3A can be substituted with tubes having similar trans-conductance with minimal (if any) sonic penalty. I have tried soviet triode strapped 6Z9P, 6Z11P, 6Z49P, 6E5P and they all sounded fine. All that for just a small fraction of the D3A price. This was not the case in common-cathode stage as all of those tubes had their own sound signature with D3A being head above the competition.
It’s actually possible to not use any JFET for level shifting at all. D3A and similar high-Gm tubes have enough “grunt” to keep their cathode impedance low enough for TDA1541A output to be still happy with it. Although this comes with increased complexity as we now have to use DC servo to bias our tube properly. I measured 200mVpp voltage swing at 18mA cathode current which equals to 50Ω I/V resistance so it’s still within what I would consider acceptable.
This increased I/V resistance, together with the fact that it’s also the same impedance that drives cathode, has increased distortion as well. But not by much as above spectrum shows clearly. Sonic wise I feel this was a step backwards, but not in any significant way. More on that later.
Couple notes on safety. As these TDA1541A DAC chips are becoming more and more obsolete and approach a zone of “unobtanium”, non-fake ones are often sold for the price of gold (in weight equivalent). So destroying one in a process of your audio nirvana search is becoming more obnoxious than ever. To avoid this heart-stopping experience you should always use back-to-back diode protection for analog output. Use only “fast” diodes having minimal capacitance like 1N4151. Their influence is below my measurement repeatability threshold and I don’t feel like they add anything to the sound.
It’s also a good idea to protect +5V power supply from reverse-biasing (BAT54) and over-voltage (6V2 Zener). Both of those conditions are possible if something goes wrong in power-on sequence or there is a short to ground on your +5V referenced HV tube power supply. If not using +5V rail for bit-current return, those can be omitted.
There are plenty enough output buffer options out there. But time and time again I find that a single JFET cascoded with high voltage depletion mode mosfet and loaded with a simple CCS does this job very well. This is your system specific. If you need to drive 10m of high capacitance interconnect cable into 600Ω, then no doubt there are far better options. For all other standard intents and purposes (2M cable into 10kΩ load) above circuit is adequate enough.
And finally I would like to present my sonic evaluation for each of these I/V stages. I don’t do this very often here, but since I made those notes to my self as I was doing listening sessions, I decided to share them. They should be treated as anecdotal evidence and only applicable to my own individual system, room, perception and preferences. Voltage supe-regulators was used for all power supply’s. Fresh New-Old-Stock Siemens D3A tubes. Cascoded JFET follower as a buffer and Jentzen Silver Caps for output coupling.
Using resistor tube load:
Sounds really good with 2X oversampling. NOS lacks HF energy. Sound stage is 3 dimensional with equal wide and height. Instruments and vocals are VERY real but a little too much in front of you. Illusion of ‘I can touch them’ is there. Details are plenty enough but not fatiguing.
Using CCS tube load:
Sound stage is different. More echo and more space around instruments. Vocals and instruments are further and sound less natural. Details are good and at the same level with resistor load.
Using resistor tube load:
Not a huge difference from GND reference, because shunt super-regs does their job very well. Seems like a little loss of energy all across the spectrum. There is less 3D imaging, vocals and instruments sound less real. Details and instrument localization is still well preserved.
Using CCS tube load:
Probably better than resistor load. Again not a huge difference from GND ref. More overall energy but not as natural, less involving.
Wow! This is by far most transparent configuration out of three. Good mix between realism of resistor load and transparency of CCS load. Vocals are ‘not in your face’ but a little bit further away, in a middle of the scene, right where you want them to be. All instruments are spread around a stage and located very precise. There is no emphasis and glare around strings and cymbals which common-cathode I/V had quite a lot. Strangely NOS started to sound better then 2xOS at 88k2. All hi-res 192k material and vinyl rips are out of this world.
This might be just mood/fatigue thing, but using NOS vocals lost their realism and image is not so stable. 2x up-sampling again makes things better. Generally more spacious sound but lacks involvement. Strings and cymbals have more spark. This could be attributed to a little bit more 2nd harmonic (0.02% vs 0.04%). Still very low. Hi-res material are not that impressive anymore. Still better then CD, but margin is not that high. Again, could be just tired ears… Next day impressions is still the same.