Valve DAC from Linear Audio volume 13

It indeed uses sigma-delta (or delta-sigma, that's the same) modulation. In fact you can choose between three sigma-delta algorithms, a seventh-order chaotic sigma-delta and two fifth-order sigma-deltas with a kind of embedded pulse width modulator with randomly rotated pattern. These last two have the advantage that they are dithered according to dither theory, which eliminates the tonal behaviour that many sigma-delta modulators have. You can also do a blind ABX test, I've included a state machine for that.

The complete design database including Gerber files, KiCAD database and FPGA files (source code and synthesized + implemented) can be downloaded for free from the downloads section of the Linear Audio website, Downloads | Linear Audio. You have to scroll down to "Design data package for Marcel van de Gevel's Valve DAC from Vol 13".

The article that explains the ideas behind the design can also be downloaded, but that will cost you almost half a euro, see Linear Audio | your tech audio resource

I rather like buying things that are not made in China every now and then, but of course you can have the PCB manufactured in China or wherever you like. Some component values may have to be adapted when the prepreg thickness is very different from the 360 um that Eurocircuits uses, but that can be done.
 
Last edited:
I rather like buying things that are not made in China every now and then, but of course you can have the PCB manufactured in China or wherever you like. Some component values may have to be adapted when the prepreg thickness is very different from the 360 um that Eurocircuits uses, but that can be done.

Thank you very much for this contribution!

Even if you have made the package freely available, it would not be ethical for me to even start a group buy without your agreement. I would plan on Chinese manufacture of the boards, looking for best price and also previous experience on this forum. I would start by opening a thread in the group buy forum to gauge sufficient interest. Let me know your thoughts.
 
Don't worry, I'd consider it an honour if you started a group buy for this. I would like to warn you, though, that the PCB is not the only expensive part, see also posts 8 and 9.

Yes, of course. That is the beauty of a group buy, if there is enough interest even parts other than PCB's can be sourced at lower cost. It really just depends on interest more than anything else.
 
Marcel, I am starting to get some questions back from the PCB houses, and I have some of my own:

The filter PCB gerber files do not include silkscreen or solder mask.

What about PCB specification itself: substrate, thickness, surface finish, etc.? What are they for your prototypes, can we use this, or do you have suggested improvements?

You mentioned that there is a error with the transformer solder pads. Is this because the software library assumed a different Jensen form factor? This may be important if we seek a substitute supplier such as Cinemag.

You mention that the transformer is optional. The schematics in the design package don't show that option. Is it to be found somewhere?
 
Marcel, I am starting to get some questions back from the PCB houses, and I have some of my own:

The filter PCB gerber files do not include silkscreen or solder mask.

True, the filter PCB was targeted for the Eurocircuits naked proto process which has no silkscreen or solder mask.

What about PCB specification itself: substrate, thickness, surface finish, etc.? What are they for your prototypes, can we use this, or do you have suggested improvements?

FR4, about 1.55 mm total thickness, single-layer filter board, four-layer main board with 360 um prepreg thickness. I'll look up the other data soon.

You mentioned that there is a error with the transformer solder pads. Is this because the software library assumed a different Jensen form factor? This may be important if we seek a substitute supplier such as Cinemag.

I simply made a stupid mistake entering the footprint. That mistake is fixed in the layout data on the Linear Audio site.

You mention that the transformer is optional. The schematics in the design package don't show that option. Is it to be found somewhere?

There is a pdf file in the zip with the design package that shows the various filter options.
 
FR4, about 1.55 mm total thickness, single-layer filter board, four-layer main board with 360 um prepreg thickness. I'll look up the other data soon.

Thanks! It would be appreciated if you could look up the other data as well. Always better to copy a successful prototype.

I was surprised at 1.55mm thickness. With this large a board, 4 layers, and the need to survive multiple tube socket insertions and removals, I would have thought that thicker would have been the way to go. Any comment?
 
I didn't think about that at all and just used the defaults. At Eurocircuits any change from the defaults makes it much more expensive - except one: I didn't include a backside legend to the main board, but actually adding a backside legend would have cost only 0 euros extra. I can generate a Gerber file for a backside legend if you like.

Anyway, the filter board was made in the Eurocircuits naked proto process. This process is double-sided, but I used only one layer because there was no real advantage in using both.

The main board was made in the four-layer version of their PCB proto process (so not naked). It is described here:

PCB proto – the PCB prototype service from Eurocircuits – Eurocircuits

I've attached plots of the layer build-ups. You can find them on the Eurocircuits website when you go to the price calculator.
 

Attachments

  • Eurocircuitsnakedproto1layer.png
    Eurocircuitsnakedproto1layer.png
    47.4 KB · Views: 1,019
  • Eurocircuitsproto.png
    Eurocircuitsproto.png
    40.9 KB · Views: 1,009
  • Eurocircuitsprotofig2.png
    Eurocircuitsprotofig2.png
    115.5 KB · Views: 989
Attached is a zip that only contains a Gerber file for a backside legend (silkscreen) layer for the main PCB, to be combined with the main board Gerbers on the Linear Audio website if you like to have a backside legend (which can be useful, as most of the digital components are to be mounted on the backside).
 

Attachments

  • KiCadtop-B.SilkS.gbo.zip
    31.1 KB · Views: 158
The main board was made in the four-layer version of their PCB proto process (so not naked). It is described here:

PCB proto – the PCB prototype service from Eurocircuits – Eurocircuits

I have gotten some quotes back from PCB houses.

One question I have is that if I read the screenshots correctly, the prototype PCB is 0.5 oz copper top and bottom (18 um) and 1 oz copper (35 um) middle layers?

Also, I don't have KiCAD, but if I am reading the files right, copper traces for the tubes are on the top layer? I am considering getting a quote on 2 oz copper on the tube trace layer since the currents are more substantial. Comment?
 
You have read the screenshots correctly.

In general most of the traces for the valves are on the top layer, most digital traces on the back and the inner layers are mostly shields.

The heater wiring is an exception, though: it is mostly on layer 2 with the top layer as a shield. The heater currents are the most substantial currents, in total nominally 2.7 A.

See the attachments, which are some screen shots of the layout with the ground and -300 V planes switched off. Red is layer 1, yellow is layer 2, pink is layer 3 and green is layer 4. The big yellow lines on the right and the top with F1 and F2 written in them are heater wiring (placed as F1-F2-F1 to cancel magnetic fields). Close to the valves they are connected by narrower wires, mostly on layer 2, sometimes on layer 1.

So if you want to minimize heater wiring voltage drop or self-heating, layer 2 would have to get thicker metal. The impact on transmission line impedances will be negligible, so the circuit should work fine with thicker metal for layer 2.

Changing layer 1 to 70 um thick metal would affect the transmission line impedances a bit; for a 0.25 mm wide trace (the narrowest I've used), the capacitance per unit length stays nearly the same and the inductance per unit length drops about 8 %. Again I don't expect this to harm the operation of the circuit in any way.
 

Attachments

  • buizenDACprint.png
    buizenDACprint.png
    48.5 KB · Views: 1,025
  • buizenDACprint2.png
    buizenDACprint2.png
    71 KB · Views: 1,018
Last edited:
Please find attached questions from member pinocchio who for some reason refuses to post it in this thread:

Do you have any specs you can publish on this thread? I do not want to buy documents yet. Have you auditioned this DAC, if not, can you point to some reviews?

Here's some info that you be nice to have:

- DAC max bit depth and sampling frequency
- Input interface (i2s, coax, toslink, usb, etc...)
- Audio stream supported (PCM, DSD, DoP)
- The usual analog out specs
- Where can we purchase the FPGA board?
 
There are four inputs. Three of them are either AES3 or SPDIF inputs, depending on how you connect them and what termination resistors you put on the board. The fourth is meant to be connected to a Toslink receiver, either one working on 3.3 V or on 5 V. If you need more Toslink inputs and you can obtain 3.3 V Toslink receivers, you can easily convert some of the other inputs into Toslink inputs.

The inputs accept word lengths up to and including 24 bits and sample rates up to and including 192 kHz. DSD64 via DoP is also supported.

There is no USB or I2S input, but all required hardware is available, so if you are prepared to rewrite the FPGA code, you can add them yourself. The FPGA board is a TE0630 micromodule from Trenz Electronic with an XC6SLX75-3CSG484I FPGA, see Mikromodul Spartan-6 XC6SLX75-3CSG484I Industrial Temp. Range | Trenz Electronic Online Shop (EN)

If you are looking for a DAC with spectacular dynamic range numbers, my valve DAC is not what you are looking for. It is a DAC that has all analogue and mixed-signal processing done by valves. To keep the number of valves down to a reasonable number, this means that the actual DAC circuit is relatively simple, which results in relatively modest dynamic range numbers; I measured 85.76 dB(A) on the left and 91.29 dB(A) on the right channel. Nonetheless, I've never heard its noise during normal use, neither over loudspeakers nor over headphones. The harmonic distortion is about 0.016 %.

I measured the frequency response to be flat to within +0 dB/-0.21 dB between 20 Hz and 20 kHz. I don't know how much of this roll-off is actually due to the measuring equipment. By design the -3 dB point should be 82 kHz (for high sample rate PCM).

The digital filter chain has very low ripples to keep pre-echoes down. The worst digital filter is the one in the SRC4392 asynchronous sample rate converter with its +/- 0.007 dB ripple, everything else is in the +/- a few millionth of a dB ballpark.

The DAC has an apodizing filter that can be turned on or off. There is also a surprise mode for blind testing, so you can check what sounds best without having to worry about expectation bias.

Unusual features of the digital signal processing part are the fact that there is adequate headroom for filter overshoots and that the anti-imaging filter is actually designed to prevent imaging - unlike most commercial digital filters, which have no (documented) headroom and hardly suppress imaging at all between 0.5 fs and 0.55 fs.

Jitter of the incoming signal is strongly suppressed by the SRC4392, so it can work well with sources of which the clock is not very clean.
 
I just put my name into the group buy, but this project would really only be useful to me if it has a I2S input.

Wondering how hard that would be to implement? I know that's a difficult question to answer, but would appreciate any insight you could add here.

I would guess an I2S input would increase the popularity quite a bit, as there are lots of USB to I2S converters floating around these days.

Randy
 
If the only input you need is one I2S input running in slave mode with a word clock of 1/64th of the bit clock, then it's easy, because that's the format used for the audio stream from the SPDIF/AES3 interface chip (DIX4192) to the FPGA module. It should then be sufficient to leave out the DIX4192 and connect the I2S source to the wires that the DIX4192's outputs would normally be connected to. A simple change in the user constraint file could move those connections to the extension connector.

If you need an I2S interface that can work in slave mode with any sensible ratio of word clock to bit clock, then it gets somewhat more complicated because the code that's used for the connection with the DIX4192 is then not usable.

If you need an I2S interface that can work in master mode while supporting all normal audio sample rates, then it becomes a quite drastic redesign, especially if it also has to work in slave mode or if you want to combine it with SPDIF, AES3 or Toslink inputs.