By the way, the trimming range update is not included in the list:
R14, R25, R26, R27, R103, R104, R105 and R106 become 2 kohm +/- 1 %, 0.6 W metal film as otherwise the trimming range can be too small
RV1 and RV2 either stay as is (Bourns 3386P-1-502LF) or become Bourns 3296Y-1-502LF, depending on whether one prefers single or multiturn
If the trimming range should still be too small, pulling out E88CCs and putting them back in in a different order can help.
R14, R25, R26, R27, R103, R104, R105 and R106 become 2 kohm +/- 1 %, 0.6 W metal film as otherwise the trimming range can be too small
RV1 and RV2 either stay as is (Bourns 3386P-1-502LF) or become Bourns 3296Y-1-502LF, depending on whether one prefers single or multiturn
If the trimming range should still be too small, pulling out E88CCs and putting them back in in a different order can help.
Good question. I just used whatever I had lying around and only made sure that the two EF80's were from the same brand, supplier and age...
Anyway, I think it would be useful to have reasonably well-matched EF80's (U15 and U16). For the rest, the circuit should only depend on matching between the halves of a double triode, not between one double triode and another, so there should not be any need to match the E88CC's.
Anyway, I think it would be useful to have reasonably well-matched EF80's (U15 and U16). For the rest, the circuit should only depend on matching between the halves of a double triode, not between one double triode and another, so there should not be any need to match the E88CC's.
I am making the spreadsheet and had a question regarding the neon lamp. I noticed in post #14, you said the neon lamps could be replaced with LEDs and the appropriate resistors. What resistors (approximately) would be needed to make that conversion? When I used a resistor calculator, it said I needed resistors of almost 8 watts.
The neon lamps are driven from 550 uA current sources, R123 is only there for safety. If you have LEDs that light up brightly enough at 550 uA, and many modern LEDs do, then you can just replace the neon lamps with LEDs and leave everything else as is.
However, you could also simplify the circuit by removing Q1...Q7, R58, R60, R66, R68, R97, R115, R117 and R123, connecting the LEDs straight between R59, R61, R67, R93, R98, R116, R121 and ground (NOT -300 V but ground!), and reducing R59, R61, R67, R93, R98, R116, R121 to about 1 kohm.
However, you could also simplify the circuit by removing Q1...Q7, R58, R60, R66, R68, R97, R115, R117 and R123, connecting the LEDs straight between R59, R61, R67, R93, R98, R116, R121 and ground (NOT -300 V but ground!), and reducing R59, R61, R67, R93, R98, R116, R121 to about 1 kohm.
Reconstruction filter types
Marcel, do you have practical experience of the audibility of different output filter types?
In the original Valve Dac article you present a 5th order linear phase output filter with 0.05 degrees of phase shift and corner frequency of 82kHz.
In our project you suggested Butterworth 3rd order at 45kHz.
Plain old Butterworth is surely more robust as to parts tolerances and makes life easier. But is the added effort to get exact values (by combining cap values or by unwinding the coils of readymade inductors to the specific values) to make up a Bessel type filter worth it?
My guess is that a 5th order Butterworth with a corner frequency of 80kHz would have small enough phase shift below 20kHz - but who knows without having listened?
Interested to hear your opinion, or Ray's, if you already tried things out.
Marcel, do you have practical experience of the audibility of different output filter types?
In the original Valve Dac article you present a 5th order linear phase output filter with 0.05 degrees of phase shift and corner frequency of 82kHz.
In our project you suggested Butterworth 3rd order at 45kHz.
Plain old Butterworth is surely more robust as to parts tolerances and makes life easier. But is the added effort to get exact values (by combining cap values or by unwinding the coils of readymade inductors to the specific values) to make up a Bessel type filter worth it?
My guess is that a 5th order Butterworth with a corner frequency of 80kHz would have small enough phase shift below 20kHz - but who knows without having listened?
Interested to hear your opinion, or Ray's, if you already tried things out.
No, I haven't. I have only tried the 0.05 degrees linear phase filter for the original valve DAC and a slightly tweaked Gaussian-to-6 dB filter for the solid state version.
By the way, that 0.05 degrees is not the phase shift but the theoretical deviation from linear phase; phase shifts that are directly proportional to frequency don't do any harm at all as they correspond to a pure time delay. Taking into account round-off errors of the component values, it's actually closer to 0.5 degrees than to 0.05 degrees.
For the original valve DAC, I could easily compensate for the roll-off of the magnitude of the filter's response up to 20 kHz by slightly changing the coefficients of the interpolation filters. It therefore seemed to make sense to use a filter with a nice phase response and just compensate for whatever passband magnitude response it had. The drawback of this approach was the poor suppression of a 0.05 degrees linear phase filter in the first few octaves above the cut-off frequency (Bessel would have been even worse). The Gaussian-to-6 dB filters are much better in that respect.
For the raw DSD version, there is no way to compensate for the roll-off in the digital signal processing. In general the magnitude response is considered more important than the phase response, so I suggested Butterworth filters. The phase shift of a 45 kHz Butterworth low-pass filter is actually still pretty close to linear up to 20 kHz: less than +/- 0.53 degrees error from a 7.283 us pure delay for third order and less than +/- 0.6 degrees error from an 11.686 us delay for fifth order.
By the way, that 0.05 degrees is not the phase shift but the theoretical deviation from linear phase; phase shifts that are directly proportional to frequency don't do any harm at all as they correspond to a pure time delay. Taking into account round-off errors of the component values, it's actually closer to 0.5 degrees than to 0.05 degrees.
For the original valve DAC, I could easily compensate for the roll-off of the magnitude of the filter's response up to 20 kHz by slightly changing the coefficients of the interpolation filters. It therefore seemed to make sense to use a filter with a nice phase response and just compensate for whatever passband magnitude response it had. The drawback of this approach was the poor suppression of a 0.05 degrees linear phase filter in the first few octaves above the cut-off frequency (Bessel would have been even worse). The Gaussian-to-6 dB filters are much better in that respect.
For the raw DSD version, there is no way to compensate for the roll-off in the digital signal processing. In general the magnitude response is considered more important than the phase response, so I suggested Butterworth filters. The phase shift of a 45 kHz Butterworth low-pass filter is actually still pretty close to linear up to 20 kHz: less than +/- 0.53 degrees error from a 7.283 us pure delay for third order and less than +/- 0.6 degrees error from an 11.686 us delay for fifth order.
Thank you for your explanation - I will go with the easier to implement 5th order Butterworth then.
This web calculator I found quite handy:
http://www.wa4dsy.net/filter/filterdesign.html
You can tweak impedance or corner frequency until you end up with a mH value for the inductors which is available ready-made.
Cross checking in LT Spice or Micro-Cap you can estimate which deviation in magnitude and phase is entailed when rounding out the specific cap values to available ones. Not too severe, I find.
This web calculator I found quite handy:
http://www.wa4dsy.net/filter/filterdesign.html
You can tweak impedance or corner frequency until you end up with a mH value for the inductors which is available ready-made.
Cross checking in LT Spice or Micro-Cap you can estimate which deviation in magnitude and phase is entailed when rounding out the specific cap values to available ones. Not too severe, I find.
It looks like that calculator is exclusively for filters with equal terminations on the input and output sides, so you will have to split the termination such that that becomes true.
When you put 866 ohm to ground after the 10 uF DC blocking capacitors, you get an impedance to ground of about 750 ohm including the effect of the anode resistors. You can then also put resistors of 750 ohm or a bit more at the output side of the filter: for example, 768 ohm will be the correct value when the amplifier's input impedance is 32 kohm. The filter can then be designed for 750 ohm termination impedance.
LC filters with equal terminations on both sides have the reputation of being relatively insensitive to component tolerances, particularly in their passband, so that's a clear advantage.
To date I've not explored alternative output filter arrangements so I'm still using the 3rd order filter that Marcel designed for the DSD-only Valve DAC.
Ray
You can connect the transformer output to either a balanced or an unbalanced input, but doing both simultaneously complicates things a bit. If you are willing to accept a halved signal level at the unbalanced output (so about 0.5 V RMS maximum level) and a non-floating balanced output, you can do this:
-Ground the centre tap on the secondary side (Jensen JT-11SSP-7MPC pins 12 and 7 or Lundahl LL1684 pins 10 and 9 for the raw DSD version)
-Take the unbalanced signal from the positive output (JT-11SSP-7MPC pin 8 or LL1684 pin 6)
-Connect the balanced output normally (JT-11SSP-7MPC pins 8 and 11 or LL1684 pins 6 and 7), also connect the secondary resistor or RC series network normally
I just use an XLR to cinch cable to connect the output to the unbalanced input of my preamplifier: XLR pin 2 to the centre of the cinch connector, pins 1 and 3 to the shield.
Edit after seeing your update:
Don't ground the secondary centre tap, just connect pins 8 and 11 of the JT-11SSP-7MPC to either the centre and shield of the cinch connector, or to pins 2 and 3 of the XLR - or use a home-made XLR-to-cinch cable.
-Ground the centre tap on the secondary side (Jensen JT-11SSP-7MPC pins 12 and 7 or Lundahl LL1684 pins 10 and 9 for the raw DSD version)
-Take the unbalanced signal from the positive output (JT-11SSP-7MPC pin 8 or LL1684 pin 6)
-Connect the balanced output normally (JT-11SSP-7MPC pins 8 and 11 or LL1684 pins 6 and 7), also connect the secondary resistor or RC series network normally
I just use an XLR to cinch cable to connect the output to the unbalanced input of my preamplifier: XLR pin 2 to the centre of the cinch connector, pins 1 and 3 to the shield.
Edit after seeing your update:
Don't ground the secondary centre tap, just connect pins 8 and 11 of the JT-11SSP-7MPC to either the centre and shield of the cinch connector, or to pins 2 and 3 of the XLR - or use a home-made XLR-to-cinch cable.
Last edited:
I've finished a parts list for the main PCB and attached the spreadsheet and linked the Mouser Cart. This list omits some PSU related stuff because I'm using off-board solutions.
Mouser Electronics
Mouser Electronics
You may already know this, but if you want to program the FPGA yourself, you also need a JTAG interface. I use this model: JTAG HS2 Programmierkabel | Trenz Electronic GmbH Online Shop (DE)