I believe that the interelectrodes capacitances of the tubes plays a very important role here, due to high speed switching. E88CC is one of the best dual triodes for this purpose (the balance of capacitance and gm). I believe that any tube with even higher gm that have relatively low capacitances will work here (eg. 6FQ7 is no-no). But is important to note (if I understand well) about the balance between the each one, and again, the E88CC is good for finding reasonably matched/not so different inner (oops, internal) triodes, for not needing a too heavy anode adjusting, and not achieving too good conversion.
One reason was that I happened to have some, but that was not the most important consideration.
You need a valve that is fast, like @DIYBras already explained, that doesn't need too much anode-cathode voltage, as there are two layers stacked on top of each other and you lose some 90 V across the tail resistors of the bottom differential pair, and you need to be able to properly switch a differential pair with just over 5 V grid voltage difference. The E88CC met these requirements, but it could very well be that other high-transconductance valves do the same.
Regarding mu, when I built the DAC, I naively thought it didn't matter much as long as the valves switch on and off, but in retrospect, I expect that the noise modulation gets smaller with higher mu.
If you want to try something like that, make sure to keep the extra wiring as short as you possibly can. The parasitics will detune the clock tank (for the full valve DAC, that is) and worsen data-to-clock and data-to-reference crosstalk.
Just about valve subbing... I have a several 5651WA, that have lower operating current than the 85A2 (that I have only one...). I can adjust the reference current to i'ts recommended value (2.5mA), but I wonder about the internal noite. In theory can be a little higher, since it have higher internal dynamic resistance. But selecting the quieter ones, or perhaps filtering the valve itself (the classic 10nF in parallel with it), but it will cause a false trigger for the "ignited85A2". Then ocurred to me to put some capacitor in parallel with the R127 to mitigate it, but then I don't know if the resulting low risetime will upset the "ignited85A2" reading in the FPGA.
The higher internal resistance also suggests to use a regulated or active filtered -300V, or making a CCS, adding a resistor for Q8 emmiter, and adjusting R126 for proper operation, and not reducing too much the R126, for not causing false triggering of the "ignited85A2".
All of this, for trying to use parts that I have when possible, without losing performance.
Or, for now I use my only 85A2. It are now used; since 2010 are installed in a bench PSU. I not used it for so much hours, so perhaps it will be healthy yet...
The higher internal resistance also suggests to use a regulated or active filtered -300V, or making a CCS, adding a resistor for Q8 emmiter, and adjusting R126 for proper operation, and not reducing too much the R126, for not causing false triggering of the "ignited85A2".
All of this, for trying to use parts that I have when possible, without losing performance.
Or, for now I use my only 85A2. It are now used; since 2010 are installed in a bench PSU. I not used it for so much hours, so perhaps it will be healthy yet...
I think I would simply change two resistors to R128 = 82 kohm, 1 W or 2 W, R127 = 470 ohm. Comparing the datasheets, I see that the 5651 is pin compatible with the 85A2 and has the same reference voltage, as you already knew.
The white noise of the 5651 will be slightly worse than that of the 85A2 because of its lower operating current, but the ratio of the white noise to the reference voltage is still very good compared to almost all bandgap references used in solid-state DACs. Besides, the noise is still filtered by R131 and C118.
The very low frequency noise that passes R131 and C118 will be mostly 1/f noise. I don't know whether the 5651's 1/f noise is somewhat better or somewhat worse than that of an 85A2 or whether it depends on the brand or on exemplaric spread. What I do know is that the 1/f corner frequencies I saw in the literature about glow discharge reference valves were pretty good.
The white noise of the 5651 will be slightly worse than that of the 85A2 because of its lower operating current, but the ratio of the white noise to the reference voltage is still very good compared to almost all bandgap references used in solid-state DACs. Besides, the noise is still filtered by R131 and C118.
The very low frequency noise that passes R131 and C118 will be mostly 1/f noise. I don't know whether the 5651's 1/f noise is somewhat better or somewhat worse than that of an 85A2 or whether it depends on the brand or on exemplaric spread. What I do know is that the 1/f corner frequencies I saw in the literature about glow discharge reference valves were pretty good.
Sounds very good. I will make it.
In another application, I have a SPUD amplifier with 12AL11, and I regulated the g2 voltage of input voltage amplifier stage using a 5651, only decoupled with the 10nF cap, with great result (low noise).
In another application, I have a SPUD amplifier with 12AL11, and I regulated the g2 voltage of input voltage amplifier stage using a 5651, only decoupled with the 10nF cap, with great result (low noise).
Another thing I want to make is using DC for the heaters. Although I see that the main work of tubes here is with "carrier" at RF rate, is better safe than sorry, due to very large noise that sometimes the mains have here.
Slowly reading the thread when I have time...
Excellent to know! Eager to know the results 🙂
Thank you for the guidance. Another question/thought was whether the resistor loading could be substituted with a choke on the tail/cathode, and an interstage on the anode. Not sure if that would work or be a good idea; obviously adds cost. For the interstage (perhasp 1+1:1+1) I was thinking two primary windings with the balance trim pot between them, and the filter + i/v on the secondary. That way no coupling capacitors would be needed.
How do you want to define the tail current then? It's now (mostly) set by the reference voltage and the tail resistor.
Hi, can someone please point me to an explanation of the schematic of the dsd-only version? I would not use tubes if possible. What adjustmment are needed for dsd128 vs 256 vs 512? Anyone has experiences better results at a certain dsd?
I will use HQPlayer for upsampling to dsd.
Thanks in advance,
Roberto
I will use HQPlayer for upsampling to dsd.
Thanks in advance,
Roberto
Roberto,
The latest and possibly the best of Marcel's DSD only dacs is the one at: https://www.diyaudio.com/community/threads/return-to-zero-shift-register-firdac.379406/ No tubes in that one.
Also if any interest, some accessory boards which can be used with the dac are at: https://www.diyaudio.com/community/threads/general-purpose-dac-clock-board.413001/
Best,
Mark
The latest and possibly the best of Marcel's DSD only dacs is the one at: https://www.diyaudio.com/community/threads/return-to-zero-shift-register-firdac.379406/ No tubes in that one.
Also if any interest, some accessory boards which can be used with the dac are at: https://www.diyaudio.com/community/threads/general-purpose-dac-clock-board.413001/
Best,
Mark
My main application for the valve DAC will be in place of a DAC using ES9018K2M in passive mode and all default mode (S-PDIF difrect input to the ES9018), using one DIYINHK board, feeding my DIY Stax headphones amplifier ("earspeakers"). The output in passive mode is a little less than the Valve DAC, so will be good for that, also. I will not use the DAC output trafo just because the amp requeriments are for true balanced input.
The phones amp uses EF184 for first stage (balanced) and output with 4P1L in push-pull. This resulted in a absolutely inaudible noise for the phones, and very low distortion; both tubes/valves are very linear for such usage.
I remember that Marcel mentions the Valve DAC not being a add-on distortion box, but being a most low distortion possible with it's concept; will be very interesting to use just with Marcel's low distortion valve application!
I don't know how the digital filtering operates in the ES9018 at default mode; I don't have the datasheet, due to secrecy about the ESS. I believe that not being it's best. Same for the S-PDIF input. Although it sounds good in this system.
But I bet that the Marcel custom dedicated filters being better than the default ones in the ES9018...
PS.: edited due to bad spelling
The phones amp uses EF184 for first stage (balanced) and output with 4P1L in push-pull. This resulted in a absolutely inaudible noise for the phones, and very low distortion; both tubes/valves are very linear for such usage.
I remember that Marcel mentions the Valve DAC not being a add-on distortion box, but being a most low distortion possible with it's concept; will be very interesting to use just with Marcel's low distortion valve application!
I don't know how the digital filtering operates in the ES9018 at default mode; I don't have the datasheet, due to secrecy about the ESS. I believe that not being it's best. Same for the S-PDIF input. Although it sounds good in this system.
But I bet that the Marcel custom dedicated filters being better than the default ones in the ES9018...
PS.: edited due to bad spelling
After reading the Marcel's Valve DAC article, I'm payng a little more attention to the digital filtering and it's specs. Well, even some time before, coincidentally, I'm started something for playing with it: the already-mentioned DAC NOS (this with TDA1541A); NOS for playing with digital filters in the PC, like the PGGB-RT (https://www.remastero.com/foo-pggb-rt-guide.html), it's filtering uses the PC computing power for a "exaggerated" specs. Recently I'm finished it. Then this Valve DAC have a choice of digital filters. So, this is my time for playing with various filter concepts (and DAC types)!
Of course, my assembly of the Valve DAC is, like I said before, for the hobbyst sake of being final demodulated by valves (like those hobbystic things; a 2024 DAC with the ancient1541...). The filter and sigma-delta choices are a interesting plus.
Of course, my assembly of the Valve DAC is, like I said before, for the hobbyst sake of being final demodulated by valves (like those hobbystic things; a 2024 DAC with the ancient1541...). The filter and sigma-delta choices are a interesting plus.
OOPS: edited:
According with datasheet, is not so different for specs than eg. the DF1704 I use for my PCM1704 DAC (eg. 0.003dB vs 0.005dB passband ripple).
I forget TWO zeros for the DF1704:
According with datasheet, differs for specs than eg. the DF1704 I use for my PCM1704 DAC (eg. 0.003dB vs 0.00005dB passband ripple). Then the DF potentially have less pre-echo than the stock ES9018.
Last edited:
Disclaimer:
Oops, since I partially load the output of the ES, the output is less than if we use it completely unloaded, so is the motive for having less signal output that the Valve DAC.