Hmmm, forgive me, but the second stage without degeneration in a high feedback application, is that really that great an idea? Cascoded or not.
I see things here, that from a classic audio design background, raise alarm bells. But perhaps I am completely wrong. To me it screams SID/TIM/stiction.
Thor
Yesterday I looked at the proposed Marcel circuit only in AC mode and at its noise production.
Today I also simulated in Transient mode and found that the amp was oscillating.
This was caused by the part at the right, responsible for shifting the output signal to around zero volt.
In the upper right corner of Marcel's proposal in #3304, there is a 4.7nF cap inserted on one side of the current mirror.
Increasing or decreasing it's value didn't bring a solution,
So, I had to remove this one and add a 100nF cap to the other side of the current mirror, called "B" in Marcel's sketch.
Also the 22uH at the input pair could be discarded, at least with the ZTX851's that I used, because I was not successful in finding correct spice models for the BC639.
I made a subcircuit from the proposed amp and called it "Marcel-01" see first attachment.
In the second attachment the subcircuit is inserted in the complete filter design, driven by the two Firdacs represented as voltage sources with a serial resistance of 378R.
In the second attachment also the FR is shown, both SE and Diff for 0dB input, directly showing the overall gain.
The third attachment shows the CMRR on the SE and Diff outputs for a 0dB signal on both inputs.
The CMRR in Diff mode is excellent, but in SE has worsened at higher frequencies because of the the added 100nF cap that was needed for stability.
Maybe there is an alternative solution to get the amp stable that does not affect CMMR in SE that much..
Hans
Today I also simulated in Transient mode and found that the amp was oscillating.
This was caused by the part at the right, responsible for shifting the output signal to around zero volt.
In the upper right corner of Marcel's proposal in #3304, there is a 4.7nF cap inserted on one side of the current mirror.
Increasing or decreasing it's value didn't bring a solution,
So, I had to remove this one and add a 100nF cap to the other side of the current mirror, called "B" in Marcel's sketch.
Also the 22uH at the input pair could be discarded, at least with the ZTX851's that I used, because I was not successful in finding correct spice models for the BC639.
I made a subcircuit from the proposed amp and called it "Marcel-01" see first attachment.
In the second attachment the subcircuit is inserted in the complete filter design, driven by the two Firdacs represented as voltage sources with a serial resistance of 378R.
In the second attachment also the FR is shown, both SE and Diff for 0dB input, directly showing the overall gain.
The third attachment shows the CMRR on the SE and Diff outputs for a 0dB signal on both inputs.
The CMRR in Diff mode is excellent, but in SE has worsened at higher frequencies because of the the added 100nF cap that was needed for stability.
Maybe there is an alternative solution to get the amp stable that does not affect CMMR in SE that much..
Hans
Attachments
I disagree. Theory suggests that some potential signal fidelity impairments can be reduced in a balanced design but that needs great care.
A balanced circuit realised to the same degree of care as an SE circuit will have 3dB more noise and n3dB more H3 and odd order distortion. In Theory. So Theory actually states "Balanced is worse" in some areas of signal fidelity.
Humans like symmetry. Balanced is symmetric.
It does not go against theory. Instead most "engineers" do not understand the theory and instead believe in hearsay and old wives (ok, old engineers) tales.
Now In Marcel's DAC there are many good reasons for him to choose balanced. It helps the digital side, loading on the analogue supplies etc.
But you now have the difficulty that you MUST use the balanced out (why not a Transformer in zero field configuration?) or you you risk disturbing the balanced setup, creating another old engineers tale of "balanced is mo betta".
Thor
The picture would be more complete when added that In Balanced mode the signal increases by 6dB, resulting in a S/N improvement of 3dB, still with the same percentagr odd order HD and much lower even order distortion.
What it does to sound perception is a different story.
The truth is that we still cannot make a 100% relation between logic and sound perception.
Hans
That's the theory. Despite the above, try SE anyway into a 10k shielded line transformer load (with DC blocking). There is very little loading on the dac to unbalance it. Maybe it can be balanced more with a 10k resistor on the inverting side. Once the clocking is fixed, and once the Vref power is a little better (with some resistor loading), the dac is starting to sound very good. VERY good. Not quite great yet, but within shooting distance.
Beyond that, tried balanced transformer output with this dac and with Andrea's balanced DSD dac. SE still wins for both balanced dacs. Also tried loading dacs into lower load impedances. In some cases it can help some, or it can make things a little worse.
Last edited:
Few years back I built a NOS AD1862 dac which was quite similar to Miro's dac. As I'm not a huge fan of NOS I have since made a new dac that uses PMD100 filter which IMO is an improvement over the NOS dac. Audible differences between well made dacs whether they are R2R, DS, RTZ, NOS or OS are small. And when it comes to DSD the sound is quite dependent on the digital filter and DSD modulator and not only dac. But most importantly preferences are subjective. So I suggest you build one and decide yourself. But please note that Marcel's RTZ dac is a much more complicated build than Miro's dac.
Do you mean Miro's dac and Bruno Putzeys' Mola dacs sound only slightly different?
Its hard to make sense out of how DSD dac sound can be quite dependent on modulators and filters, yet no matter they are still only small differences compared to any well made dac?
I mean, in a sense is probably true that all well made dacs reproduce the melody and the tempo recognizably about the same? However, I can say that my Topping D90 (AK4499 version) does not compare that well with my Andrea DSD dac. Not on electrostatic speakers anyway.
Last edited:
There is too little information here.
As in balanced the transformer see double the level, we will see more low frequency odd order harmonics.
There are too many variables being changed at the same time.
You should use two identical transformers and then connect the secondaries in series to get summing.
Or as I suggested, use a zero field setup to dramatically reduce transformer distortion. No need for op-amp's, you can use Tubes.
Here a principle design of a Mix-Bus with a zero field transformer (Gyraf G-16 Active Summing Mixer):
The DAC's current and/or resistors form the input resistors of the Zero Field Circuit.
The tube stage can be anything you like, you could (for arguments sake) do a 12AX7 with 12AU7 Cathode follower instead of an ECC88 SRPP.
Ideally the transformer is low DCR and for a DAC that likes to see a virtual ground, use a large step-up (like 1:10).
The transformer output operates into an active virtual ground the impedance of which is down to the OLG of the Amplifier. If we can make the impedance negative we can compensate the winding DCR.
Used like this the transformer sees very little signal levels and thus distortion is minimal, also - frequency response becomes very wideband.
Transformers are interesting devices, the loading is down to the load resistor in the mid band and parasitic at LF & HF.
Unlikely.
If you are "in the middle" all bet's are off. Either a (virtual) ground (see zero field above) or open circuit are best.
With ESS 9018 the best sounding setup I ever found was a 10K:10k mu-metal core transformer with +26dBu @ 50Hz level handling, followed by a Cathode follower (more precisely, an Alan Wright SLCF with a total of 2 double triodes and +/-200V rails, DC coupled output, "Romy Da Kat DC coupled servo Bias". Transformer connected BAL in SE out.
So Signal path:
DAC -> TX BAL in -> TX SE out -> 1 Resistor, cascoded triode as follower, cascoded triode as CCS/White follower
But that needed a heroic transformer just to handle 2V RMS from the DAC in voltage mode with low distortion.
I never tried a zero field setup, but as the TX vendor send me the wrong item and will not change it, I now have the opportunity to try it out with ES9038 on a Rod Rain DAC PCB.
Thor
Last edited:
Regarding zero-field (in the audio band), not sure if that's the issue with the transformer I have. It more like I wonder how the RF is interacting with the high-nickel core. Obviously the core is lossy at RF but the RF is also sort of like tape bias. Too much RF bias is not necessarily better.
A passive pre-filter might be worth trying. Or, even better, I would like to have a one on one with the transformer designer. But that may not happen because of human factors.
A passive pre-filter might be worth trying. Or, even better, I would like to have a one on one with the transformer designer. But that may not happen because of human factors.
Last edited:
Unless it is an insanely big permalloy core monster like the WE111C repeater coil or bigger (we are talking transformers that push a kilogram or more (WE111C is 4.5 lbs) you have an issue with that.
Level vs frequency and distortion for an extremely respectable line input transformer (10k:10k)
from:
Jensen JT-11P-1 Transformer Data
As we see, as little as +4dBu / 1.23V input voltage exceeds 0.001% THD or -100dB below 250Hz, which is the music's foundation. Due to the underlying mechanics this HD will be H3 and odd order.
HD goes up pretty linear level for any given frequency.
So for +10dBu (2.46V) distortion will double for a given frequency.
A lot more details on all this can be found here:
https://www.onallbands.com/wp-content/uploads/2019/08/Non-LinearXformerBehavior.pdf
Hmmm, I suspect compared to the main signal, the RF is not material.
You should have a pair cap's on each phase to ground ahead of the transformer. Analyse the complex impedance of the circuit first and make sure it is critically damped or add a snubber on the secondary to kill resonance's.
I find killing a resonance flat sounds a little flat, just 1-2dB ultrasonic peaking seems the sweet spot.
Thor
Transformers I have are rather unusual. Probably better if I don't say too much about that. Anyway, there is no immediate rush to solve output stage issue. If Marcel gets something working then I will probably give it a try. In the meantime I am busy enough trying to finish up on my clock board project (which now seems to be turning into a reclocker board project too). When I get the clock stuff off my plate, then will be a good time to take up other remaining issues.
The huge transistors you use for Q17 and Q18(?) probably cause the pole and the right half plane zero of this differential pair to end up too close to the origin.
BCP53 and BCX53 are supposed to be equivalent to the BC640, and BCP56 and BCX56 to the BC639. Maybe you can find Spice models for those.
If not, if LTSpice allows fractional m or mult factors (I don't remember which of the two is Spice syntax), you could give your ZTX851's an m factor of about 0.1 to 0.2.
Removing the 22 uH degrades the noise performance at high frequencies, as should be obvious from inspection of the circuit.
Well, be that as it may be, but it is all basic physics. I have specified and designed transformers for many jobs, mains, switching supies, tube outputs, line level, MC Pickup. I think I know a little bit about transformers.
The LF distortion behavior is well understood.
High nickel permalloy/mu-metal etc or high nickel metglass cores have very low midband distortion.
LF distortion (given we keep the core material fixed) is down to magnetic crossection, number of turns and signal levels at the winding.
LF distortion at a given level can be reduced by reducing the nickel content, but that raises MF/HF distortion. So a steel core transformer can handle a lot of level at low frequncies, but generally distortion at lower levels is high across the board.
It's all a compromise, but a low distortion line level transformer needs a good fraction of a kilogram of nickel. That will get you three to four zeros after the decimal point to below 100Hz.
The alternative is a smaller core and zero field circuitry. But that invariably involves feedback. But it lowers cost, size at similar performance levels.
Thor
Marcel,
Yes, you are right, adding the 22uH reduced the noise at high frequencies, so it must stay.
See attachment with SE output noise for 22u in Red and without inductor in Blue, resp. 1.0uV and 1.8uV from 20Hz to 20Khz.
I also tried the BCP56 that you mentioned instead of the ZTX851 and indeed C7 the 100nF cap could now be removed without affecting stability.
I could also put the original 4n7 cap back, but saw no difference in stability.
Because of the 100nF that's no longer there, the CMRR has also been greatly improved at high frequencies.
See second attachment.
Hans
Yes, you are right, adding the 22uH reduced the noise at high frequencies, so it must stay.
See attachment with SE output noise for 22u in Red and without inductor in Blue, resp. 1.0uV and 1.8uV from 20Hz to 20Khz.
I also tried the BCP56 that you mentioned instead of the ZTX851 and indeed C7 the 100nF cap could now be removed without affecting stability.
I could also put the original 4n7 cap back, but saw no difference in stability.
Because of the 100nF that's no longer there, the CMRR has also been greatly improved at high frequencies.
See second attachment.
Hans
Attachments
Last edited:
Thor,
Yes, of course. But there are also some details you didn't get into into, including maybe a little bit that probably won't occur to you. Even so, its all physics just as you say. The internal data format in ESS and AKM dacs is just physics too, but its a big secret to those dac chip seller guys. Similarly, the transformers guys in this case don't want me spilling some of what are to them are their big secrets. I said I wouldn't, so I won't.
What I can say is that if you want step up transformers to play around with for dac I/V, looks like Quadratic Audio is one of the companies getting some good reviews. Haven't seen any other offerings from them as of yet though.
Mark
Last edited: