Super valve common grid for DACs
I had a previous experience building a 6N8S powered pre-amp. I took directly the signal after the opamp of an enter level CD player (I don’t even remember the brandt). I must say I was quite impressed by the result. The main idea was to do quite the same thing. As soon as I submitted the idea to the Audiyoan forum lot of people reacted and my post became the new incarnation of all old threads covering DACs and how to extract signal from them.
Some of the most famous DACS for DIYers are Phillips’TDA1541 and the cheaper one TDA1543. Both are 16bits DAC and both convert digital signal to analogic current. The main problem is: we know how to amplify voltage but not current. The first step is to convert the DAC output current to a voltage signal .
The ugly passive way
The first idea most have (I had) is to use a resistor, U = R∙I so we convert the output current of the DAC to voltage. Well, understanding a bit the internals of a current DAC will help us to understand why this solution is the worst we can use.
In the DAC, each bit controls a constant current source that makes current flow or not through a resistor network. If the voltage at the output pin changes, that induces a summation error in the chip resulting in distortion. The datasheet is clear: distortion and precision are given for a voltage not changing over 25mV. With a 2mApp output, this would let us use a 10Ω resistor and obtain 20mVpp output which means amplifying it by 20dB before we can send it to the real amplifier.
This method is used by a LOT OF cheap DACS you may find on internet. Some of them just use a 1kΩ resistor as passive IV-converter which result in very high distortion. The TDA1543 is famous because using a resistor a way above the 10Ω limit produces something roughly “listenable”. Another way to lower the converter transimpedance is to increase the output current by using several TDAs in parallel. Steam engine solutions are using up to 8 TDA to lower the impedance down to 125Ω ... still more than 10 times over the limit.
The passive way has no really interesting results comparing to most of CD player. Either you choose to stay in the 10Ω limit and you have at least the max distortion of the DAC plus the need to amplify this distortion. Otherwise, you go far beyond the limit and I can tell you will switch off your DAC after several hours listening, your hears calling for silence.
The bad OpAmp virtual ground
The solution commonly used in (old) CD players using TDA1541 and TDA1543 is to create a virtual ground using an operational amplifier.
The current arrives on the op-amp pin but as the input impedance of an opamp is infinite, it flows through the feedback resistor. The gain of the opamp is compensating the voltage drop across the resistor. The – pin stays at the same potential while the output voltage is varying, the signal has been converted to voltage. This is a very effective circuit, the simulation using a “normal” opamp showed an input impedance down to 0,1Ω at 1kHz, 100 times lower than the 10Ω limit with a 0,3V output … still need 10dB here.
Yes but as the opamp open loop gain is dramatically dropping with the frequency, its performances too ! Even if it is lower than the limit, the transimpedance changes from a factor of 20dB or more between 20Hz and 20kHz and it may be much worse, that really depends on the opamp you are using. Of course, I do not think one might hear any difference, the very bad excuse is I think the sound travels better in vacuum than in silicon.
One opamp-like virtual ground using valves has been prototyped by Totof at Audiyofan but it could not give the hopped results. Sound was good that’s it, and it was a steam engine to tune.
The good valve common grid
This kind of circuits are also called current conveyors. They present a very low input impedance, a good transformation ratio and a high output impedance. The current input is made on the cathode. From here, the only loads seen are the cathode load in parallel with the anode load. From a cathode perspective, the anode load is seen divided roughly by the gain of the mosfet/valve. In fact the anode load seen from the cathode is approximatively 1/Gm where Gm is the transconductance of the component … very low.
Building a common grid current conveyor with valves is as easy as it is with mosfet but Gm of tubes is significantly lower (5 to 30mA.V⁻¹) than mosfet( > 200mA.V⁻¹ to several A.V⁻¹). With a high Gm valve, we can present a 1/30.e⁻³ = 34Ω input impedance circuit … which is a way worse than the previous opamp schema but it is really stable over all the audio bandwidth and with 2mApp input, we have 4Vpp output !
In the cathode here, we have set a current source composed by the BC547C transistor. This current source is set to 30mA where the 6Z43P valve has its maximum transconductance.
Couldn’t we do something with this unattended grid ?
The super common grid
If one valve is not enough, use 2 ! The goal here is to use the grid to say to the triode not to move its cathode from a mV !
What does happen here ? If the triode’s cathode moves from a single µV, this change is collected on the mosfet’s gate, amplified and inverted on its drain. The amplified-inverted change is applied to the triode’s grid which sticks the cathode to its potential. In other words, we multiply the transconductance of the triode with the mosfet’s one. The results are really good: the input impedance is 0,03Ω from 20Hz to over 100kHz, 4Vpp output and 2,2kΩ output impedance which is acceptable if we consider the following amplifier to be 50kΩ input impedance.
The upper curve is the phase and the lower is the voltage change at input for 2mApp.
It has been a long work to tune the circuit. First, we had to prevent the circuit from resonating at several MHz. The input impedance is so low, we could set a 1µF capacitor in parallel to trap all oscillations over 100kHz. Then we had to set the Lin input pin between 2V and 3.8V as specified in the datasheet. Using a mosfet for feedback really made it easier because its very high transconductance leaves room for change.
In real life, the C5 polarized capacitor is coupled with a film 4.7µF, the C4 1µF is a MKP soldered on the gate’s resistor (not represented here). The QTLP690C led is a cheap red led you can find everywhere. Under normal conditions with the 200V power supply you will find the following values: Vdrain = 85V, Vanode = 135V, Vcathode = 3V, Icathode=30mA
We can spot here my loosy power supply doing noise up to -70dB at frequency multiple of 100Hz. The harmonics are composed of the 2nd harmonic, the third and the fifth all around -90dB. I am waiting for a new transformer to build a new power supply. The expected SNR is more than 90dB. (software used: baudline )
How does that sound ? Well, a friend and I bought a TDA1543 passive printed circuit with a DIR9001. We hacked it to use with the audiyofan converter. It sounds like every converter should sound: very clear and transparent. Strangely, one of the feelings I had when first listening to it was frustration. Of course I was pleased with transparency, details presence and the superb stereo image but something was missing like some instruments usually at the front were more “in the middle”. I am convinced today my previous CD player with oversampling technology used filters to make the sound «more beautiful». Because when you buy a CD player in a shop, it has, let’s say, 5 minutes to convince you to buy it.
This DAC+converter is now my everyday source to feed my 6P14P push pull, I will not step back, it is just the sound I was looking for.
At Audiyofan we had great time finding this circuit and it took us more than 6 months to get something working and simple. The schematic I provide is an hybrid variant and is released under the free CC by SA license which allows you to use it, publish your work, modify it as long as you mention the copyleft owners and keep your changes under the same license terms.
Version 1.1 (28th of May 2011)
After 40 days listening to the 1.0 version, I modified a bit the schematic :
Changelog: – R3 changed from 2,2kΩ to 2,7kΩ – R5 changed from 24kΩ to 27kΩ – HT changed from 190V to 220V DC – C3 changed from 100nF to 220nF – Added C10 1µF MK
Higher output, better dynamics and contrast.
Version 1.2 (30th of May 2011)
After 2 days listening to the 1.1 version, I found it a bit too aggressive.
Changelog: – R3 changed from 2,7kΩ to 2,5kΩ – Measures from the version 1.1 were wrong.
Version 1.3 (6th june 2011)
The 1.2 version is still too boomy. After trying 2.5kΩ, 2,38kΩ I am back to 2,2kΩ and this is nice this time.
Few words about the power supply
When you apply the HT on this circuit, the C3 capacitor loads with a PLOP. This PLOP drains very slowly through the high R6 resistor. During this time, the grid is at wrong potential and then, the cathode which means the TDA output and the mosfet’s gate which slows down the process.
The simulation showed apply the HT gently could minimise the PLOP and makes the system to be at right condition in approximately 10 seconds. I have used Yves’s power supply schematic (in french, google is your friend) with a 47µF ballast which raises the HT in 45 seconds, to wake up the circuit gently.
Copyleft 2008 – Yves MONMAGNON
The schematic I have used for the 1.1 version uses a 110Vrms transformer with a Latour voltage multiplier.
In real life, R13 is not a 190kΩ but a 150kΩ + 47kΩ potentiometer to adjust the HT. When in load, the RC filter output is around 240V so expect the 300Ω resistor to be really hot, I am using a russian 5W monster that can resist the 40W power flash at start up.
PS: Here is Christophe’s version: full tube using a russian triode-pentode 6F12P.