It looks as though we'll be seeing some more builds coming through soon - good luck everyone and do report back. 🙂
If you wish to reduce noise modulation, you can reduce the impedance loading U1/U2 or try devices with higher voltage gain, such as pentodes (partition noise will be negligible because of the high signal levels).
From section 13.1 of the pdf referenced earlier. Are there any feasible ways to "reduce the impedance loading U1/U2"?
In the mean time I can wind inductors I suppose. And pull my hair out trying to select output caps...
I bought a batch of NOS Audiotronics 10uF 400V polyprop caps for the outputs in my build - very pleased with the DAC sound with them installed and good value. I'm on the look out for some more for when I build up the second PCB I had for myself from the last GB.
I saw those on your first build, they look awesome. I was looking at some Jantzen Silver Z-cap, about $50 each. Don't really want to go to crazy with it. People recommended Jupiter, but those are way too rich for my blood lol
From section 13.1 of the pdf referenced earlier. Are there any feasible ways to "reduce the impedance loading U1/U2"?
I can't think of any simple ones without side effects:
The simplest is to reduce the impedance level of the reconstruction filter, preferably also increase the 10 uF capacitors further and accept a reduced output signal level. You could make up for it to some extent by connecting the output transformer as a 1:2 transformer.
A more complicated option is to put cascodes on top of the DAC cores. You then need a more negative B- supply to make voltage headroom for the cascodes, shift several DC bias voltages, and the DC grid voltages of the cascodes have to be made trimmable so you can trim for minimum noise (like is now done with the anode resistors). You would have to determine experimentally whether there should or should not be any filtering before the cascodes.
You could also reduce the impedance level of the output filter and terminate it with resistors in series with transimpedance amplifiers or common-gate stages.
In any case, of the people who listened to the DAC, no-one has complained about this effect, and it is very benign compared to what you have with a professional tape recorder with noise reduction.
Last edited:
Thanks Marcel. You are very helpful.
I found the relevant words in the first post and have made some calculations:
I sort of indicated it earlier that I could be flexible on the output level, the filter construction and the output transformer, but perhaps not in the ways that were immediately evident. Using the method described by you above I worked with a 600ohm filter termination, the 47kR input resistor on the preamp and the 11.2kR anode resistance of the E88CC pair and have ended up with 1Vrms output before transformer losses...which is perfectly fine...that is the same level as my phonostage. Should be a little better for in-band noise which I really hope to optimise considering my 110dB/w/m sensitive horn system and my demand for "no noise". System gain structure should be very good at 1Vrms output.
At this point calculating these reconstruction filters is beyond me, but to my mind the 1:1 output transformer primary could be wound to provide a 600ohm load for the filter thus eliminating the terminating resistor. Would also be nice to get the cap values down to a size for air-caps (<1.4nF) but I have no idea if that is possible.
I found the relevant words in the first post and have made some calculations:
Output signal level:
Neglecting transformer losses, if any, the peak output voltage is about 2.5 mA times the parallel connection of three things: the anode resistors of the upper E88CCs, together about 11.2 kohm, the termination resistor, and whatever load you get from the amplifier that's connected to the output. With an 806 ohm termination resistor and a 10 kohm amplifier input resistance, that's about 1.75 V peak or 1.24 V RMS. (This is assuming that the percentage of ones in the DSD signal varies between 25 % and 75 %, as it should according to the Scarlet Book standard.)
I do not recommend using higher impedances to boost the signal level, but using lower termination impedances to reduce the level is very well possible, provided that the reconstruction filter is redesigned accordingly and the signal transformer (if any) can handle it. If you really need a higher signal level, you can use a transformer to step it up, see the left schematic of post #837 and the text of that post.
I sort of indicated it earlier that I could be flexible on the output level, the filter construction and the output transformer, but perhaps not in the ways that were immediately evident. Using the method described by you above I worked with a 600ohm filter termination, the 47kR input resistor on the preamp and the 11.2kR anode resistance of the E88CC pair and have ended up with 1Vrms output before transformer losses...which is perfectly fine...that is the same level as my phonostage. Should be a little better for in-band noise which I really hope to optimise considering my 110dB/w/m sensitive horn system and my demand for "no noise". System gain structure should be very good at 1Vrms output.
At this point calculating these reconstruction filters is beyond me, but to my mind the 1:1 output transformer primary could be wound to provide a 600ohm load for the filter thus eliminating the terminating resistor. Would also be nice to get the cap values down to a size for air-caps (<1.4nF) but I have no idea if that is possible.
preferably also increase the 10 uF capacitors further
If 250V output caps are suitable, I could probably go to 18uF
FYI, there are lots of 20pcs of Solen 10uf/400V caps on eBay for $50 shipped. They are excellent caps.
Last edited:
I usually use Jupiter copper caps but 10uF is only available in their 100V values. The option I am looking at is the V-cap Odam which goes up to 18uF in 250V variants but are still US$230 for a pair.
Sonny, look back through a few of my questions for the cap noise story. Marcel had written it all out on the front page previously but for me it did not sink in. Those large output coupling caps are used to keep in-band noise in the dac output under control. Smaller caps = more in-band noise in the lower frequencies. Larger caps = less in-band noise in the lower frequencies.
Also a factor for in-band noise is the output level of the dac. Looks like it is capable of 19V output into a very high impedance load but this high output brings higher levels of in-band noise which is why Marcel has been very careful to limit the output to 1.24Vrms...lower output would likely be better from a noise standpoint but not so good for the gain structure in everyones system. What I did is work backwards (my entire system is diy) from my amp that need 4V input to get really loud, my 8x gain preamp, and allow 6dB for headroom in the recordings and figured out that I only need my particular ValveDac to put out 1Vrms to fit in with my system. Any less and I will sometimes run out of volume...any more and I am wasting gain in the system. Everyones system will be different.
Saying all that, I like the idea of a larger high quality cap on the output of the ValveDac...won't be a Jupiter but the V-Caps are supposed to be decent to push the in-band noise lower in frequency (my speakers are flat in-room to 18Hz and I can hear a high volume 16Hz sine through my speakers) so the effort may well be warranted. But this means the reconstruction filter will need updating to match.
Sonny, look back through a few of my questions for the cap noise story. Marcel had written it all out on the front page previously but for me it did not sink in. Those large output coupling caps are used to keep in-band noise in the dac output under control. Smaller caps = more in-band noise in the lower frequencies. Larger caps = less in-band noise in the lower frequencies.
Also a factor for in-band noise is the output level of the dac. Looks like it is capable of 19V output into a very high impedance load but this high output brings higher levels of in-band noise which is why Marcel has been very careful to limit the output to 1.24Vrms...lower output would likely be better from a noise standpoint but not so good for the gain structure in everyones system. What I did is work backwards (my entire system is diy) from my amp that need 4V input to get really loud, my 8x gain preamp, and allow 6dB for headroom in the recordings and figured out that I only need my particular ValveDac to put out 1Vrms to fit in with my system. Any less and I will sometimes run out of volume...any more and I am wasting gain in the system. Everyones system will be different.
Saying all that, I like the idea of a larger high quality cap on the output of the ValveDac...won't be a Jupiter but the V-Caps are supposed to be decent to push the in-band noise lower in frequency (my speakers are flat in-room to 18Hz and I can hear a high volume 16Hz sine through my speakers) so the effort may well be warranted. But this means the reconstruction filter will need updating to match.
Hi Raja,
I've marked the resistors that I expect to be sensitive to 1/f noise(*) with "MF" or "metal film" on the schematic.
Regards,
Marcel
(*): I prefer to call it 1/f noise rather than current noise because you can also see the thermal noise as current noise when you choose the Norton equivalent of a noisy resistor. Then again, others may object against the term 1/f noise because its power spectral density is not necessarily exactly inversely proportional to frequency. Anyway, we mean the same.
Thank you so much, Marcel. This saves me from doing blind tweaks, heh.
About the output DC blocking capacitors, the issue is not low-frequency noise, but low-frequency programme signals modulating the noise floor a bit more than higher-frequency programme signals:
A. With the present output filter, increasing the DC blocking capacitor values to 27 uF or 33 uF would reduce the corner frequency below which the noise modulation gradually increases to 15.7 Hz or 12.9 Hz, respectively. Taking the lowest frequency in music to be 16 Hz, that means the reactance of the DC blocking capacitors will never contribute significantly to the noise modulation anymore. With 18 uF it is about 23.6 Hz, low enough for just about anything but church organs, with 10 uF it is about 42.4 Hz, still low enough for most music styles.
B. When you scale down the filter impedance to reduce noise modulation due to higher-frequency programme signals (thereby also reducing the output signal level), the DC blocking capacitor value needs to be scaled up inversely proportionally if you want to keep the corner frequency below which the noise modulation increases below 16 Hz. Hence, with halved impedance, you then need about 53 uF.
C. The capacitors should not be made greater than 330 uF to ensure they are properly charged when the mute relay (or JFETs) opens. Of course 330 uF is impractically large anyway.
D. 100 V working voltage of the DC blocking capacitors is more than enough for normal operation, but I can think of a number of a number of single-fault conditions that would lead to larger voltages across the capacitors. The worst would be a heater to cathode short in one of the upper valves. To ensure that the DC blocking capacitors then still block the voltage (as a safety feature), you need a higher working voltage. I'm not sure if 400 V is required or 250 V will do, I will get back to that later.
A. With the present output filter, increasing the DC blocking capacitor values to 27 uF or 33 uF would reduce the corner frequency below which the noise modulation gradually increases to 15.7 Hz or 12.9 Hz, respectively. Taking the lowest frequency in music to be 16 Hz, that means the reactance of the DC blocking capacitors will never contribute significantly to the noise modulation anymore. With 18 uF it is about 23.6 Hz, low enough for just about anything but church organs, with 10 uF it is about 42.4 Hz, still low enough for most music styles.
B. When you scale down the filter impedance to reduce noise modulation due to higher-frequency programme signals (thereby also reducing the output signal level), the DC blocking capacitor value needs to be scaled up inversely proportionally if you want to keep the corner frequency below which the noise modulation increases below 16 Hz. Hence, with halved impedance, you then need about 53 uF.
C. The capacitors should not be made greater than 330 uF to ensure they are properly charged when the mute relay (or JFETs) opens. Of course 330 uF is impractically large anyway.
D. 100 V working voltage of the DC blocking capacitors is more than enough for normal operation, but I can think of a number of a number of single-fault conditions that would lead to larger voltages across the capacitors. The worst would be a heater to cathode short in one of the upper valves. To ensure that the DC blocking capacitors then still block the voltage (as a safety feature), you need a higher working voltage. I'm not sure if 400 V is required or 250 V will do, I will get back to that later.
Last edited:
Regarding the filter inductors, since I'm using different cores with different turn counts, do the resistors need to be adjusted to account for the difference in resistance of the length of wire? I was also concerned with the diameter of the wire, since the closest thing to 0.6mm I can get domestically is 22awg, which is slightly thicker at ~0.64mm.
I've always neglected the wire resistance of the coils, so your smaller number of turns and thicker wire should only make the filter more accurate.
Is it difficult to calculate the filter for 600r termination and 18uF DC blocking caps? Or the standard 750R termination and the 18uF caps? I've spent a fair bit of time trying to emulate the standard filter but am having no luck getting your component values so am certainly doing something wrong. Happy to zap some $$ across for your time, assuming we decide 250V caps are fine to use.
Last edited:
Did you take into account that only the output side has a 750 ohm termination and that the input sees about 11.2 kohm?
- Home
- Source & Line
- Digital Line Level
- Valve DAC from Linear Audio volume 13