Thanks! The idea behind the pot attenuator was what I thought it would expand THD+N to best possible figures. But now I see that preset sweet spots might work better.
Well, optimizing for THD+N is one thing, but a soundcard will be useful for much more than just THD+N measurements. I use mine for all kinds of test signals and data processing, including loudspeaker measurements. An attenuator with fixed, well-defined levels is so much better to work with than a variable pot that requires level calibration each time you touch it.
Actually, my biggest concern about pots was distortion but 4 gang and CMRR seem to work fine. Other than that, based on my experience from the first version I think it's very handy. I can repeatedly set the exact dB levels just looking at the screen, never care watching the knobs. And when a DUT is inserted in the loop, I have built-in attenuators that otherwise I would have to add in the test set up, say if I need to set an amplifier at 1W output for reference and adjust digital scale to whatever level. That would call for a switch attenuator with narrow steps.
I like the design of your analogue front end with the current limiting depletion FETs.
Input fully balanced is fine, but there is something that nags me:
The balanced inputs are accessible via unbalanced BNC connectors.
Thus the screen of BNC cable is on one input leg -
from a standpoint of noise screening even a complete unscreened cable would do a better job.
Alternatively you could use a pair of BNC cables for each input,
not a smart solution imho.
My personal favourite is an XLF-F/TRS-6.3mm combo jack from Neutrik for these applications.
Mini XLR being another option.
Input fully balanced is fine, but there is something that nags me:
The balanced inputs are accessible via unbalanced BNC connectors.
Thus the screen of BNC cable is on one input leg -
from a standpoint of noise screening even a complete unscreened cable would do a better job.
Alternatively you could use a pair of BNC cables for each input,
not a smart solution imho.
My personal favourite is an XLF-F/TRS-6.3mm combo jack from Neutrik for these applications.
Mini XLR being another option.
Thanks! I decided for BNC because they do not occupy much space and they seem to be universal choice for bench instruments. So, I can use many interconnects, even scope probes. That said, I did build a set of custom interconnects with twisted screened cable -the same found in XLR. At the BNC end, only the twisted cores are connected. At the other end there will be an RCA or XLR or TRS wired accordingly. My conclusion is that it is impossible to create ground loop noise other than that coming from the DUT.
Gerber Files
I don't think there will be any changes on the schematics other than fine tuning. So, I can upload gerber files. Parts' values are marked on the silk screen. Two opamps are on the bottom side. I also upload the front panel. It's designed to fit the Behringer chassis but it can be used with anything else. Speaking about chassis, searching for eurocard standard dimensions I only found some with the front panel at the 10cm side while we need it at the 16cm. It would be nice if anyone could suggest something suitable. The PCB is designed for this and makes assembly very easy.
I don't think there will be any changes on the schematics other than fine tuning. So, I can upload gerber files. Parts' values are marked on the silk screen. Two opamps are on the bottom side. I also upload the front panel. It's designed to fit the Behringer chassis but it can be used with anything else. Speaking about chassis, searching for eurocard standard dimensions I only found some with the front panel at the 10cm side while we need it at the 16cm. It would be nice if anyone could suggest something suitable. The PCB is designed for this and makes assembly very easy.
Attachments
This is quite a special solution but I am sure it will work in that case. Scope probes not so much - using these will work if you add a gnd-switch that grounds one of the inputs. I see the problem of a universal bal/unbal interface solution like you.That said, I did build a set of custom interconnects with twisted screened cable -the same found in XLR. At the BNC end, only the twisted cores are connected. At the other end there will be an RCA or XLR or TRS wired accordingly. My conclusion is that it is impossible to create ground loop noise other than that coming from the DUT.
Using TRS-inputs we have bal TRS cables or unbal TS-cables to choose from without further trickery.
Last edited:
That would work for measuring relative THD, but IMO not when measuring absolute voltagage/noise params. I agree with @mbrennwa that a well-calibrated fixed attenuator would allow more measurements easier.I can repeatedly set the exact dB levels just looking at the screen, never care watching the knobs
Agreed. I ended with fixed attenuators with 6dB steps.
Looking sometimes at plots from the past with undocumented attenuator settings taught me to do so.
Looking sometimes at plots from the past with undocumented attenuator settings taught me to do so.
I think I understand the problem now. But how many steps would make this practical? For example, 4 steps is already 16 positions for two channels balanced. keeping it only at one channel still needs a bulky and expensive part for just a few steps. Not sure it fits this project. Perhaps it would be better to use something as external add on.
Why is that?Calibration in Linux/REW requires rebooting
With identical hardware the results must be basically identical, provided your SW chain is bitperfect. IIUC Arta has only ASIO and WDM. ASIO (e.g. ASIO4All or FlexASIO) is bitperfect, WDM is not, it goes via windows mixer.It is possible to reach full scale but the figures are mediocre...
Arta + wine on linux - by default WDM on wine goes through pulseaudio/pipewire on the host linux - not bitperfect. Wine can be configured to use alsa connector - bit perfect if hw:X device is then used in wine -> Arta.
REW on linux - the stock linux javasound connector always wraps hw devices with the plug plugin, therefore precautions must be made to ensure bit perfect output. REW also includes additional javasound connector which enumerates/uses PCM devices (e.g. those defined in ~/.asoundrc). That allows to configure a PCM device with fixed parameters (basically bypassing the plug plugin) and have a fixed bit perfect path.
No ground -bal/unbal- switch is needed with this. The inputs are differential floating. I have been using them with balanced, SE and bridged amplifiers without problems. The DUT determines if one side is grounded. The soundcard input doesn't need ground reference. Tested with all kind of amps.This is quite a special solution but I am sure it will work in that case. Scope probes not so much - using these will work if you add a gnd-switch that grounds one of the inputs. I see the problem of a universal bal/unbal interface solution like you.
Using TRS-inputs we have bal TRS cables or unbal TS-cables to choose from without further trickery.
I really don't know. I faced that in the past and once more spending the whole last weekend to calibrate the soundcard and still not sure if it's optimal.Why is that?
In any case, I'm using the drivers provided from DIYINHK, supposedly ASIO. ARTA with win7 and REW with Kubuntu. In both cases, I can see the device being recognized. Doesn't this mean that REW Is using the ASIO drivers? How can I verify? Having used REW in windows with java drivers in the past, I can already see great improvement moving it to Linux.With identical hardware the results must be basically identical, provided your SW chain is bitperfect. IIUC Arta has only ASIO and WDM. ASIO (e.g. ASIO4All or FlexASIO) is bitperfect, WDM is not, it goes via windows mixer.
Arta + wine on linux - by default WDM on wine goes through pulseaudio/pipewire on the host linux - not bitperfect. Wine can be configured to use alsa connector - bit perfect if hw:X device is then used in wine -> Arta.
REW on linux - the stock linux javasound connector always wraps hw devices with the plug plugin, therefore precautions must be made to ensure bit perfect output. REW also includes additional javasound connector which enumerates/uses PCM devices (e.g. those defined in ~/.asoundrc). That allows to configure a PCM device with fixed parameters (basically bypassing the plug plugin) and have a fixed bit perfect path.
This will not work with scope probes.No ground -bal/unbal- switch is needed with this. The inputs are differential floating. I have been using them with balanced, SE and bridged amplifiers without problems. The DUT determines if one side is grounded. The soundcard input doesn't need ground reference. Tested with all kind of amps.
I guess that's because frequency compensation won't work, correct? Might work with cap/resistors values adjusted for balanced input?
That depends what input device was selected in your Arta.In any case, I'm using the drivers provided from DIYINHK, supposedly ASIO
IIUC it's a standard UAC2 device, no need for any extra driver in linux (and likely in windows non-asio either).n both cases, I can see the device being recognized.
There is no ASIO on linux.Doesn't this mean that REW Is using the ASIO drivers?
By what device you chose in your REW config.How can I verify?
REW with wasapi exclusive on windows will yield identical results to REW with plughw (without any automatic conversion) or properly configured PCM device on linux. In all these cases the samples go directly to the device, via a bitperfect driver.Having used REW in windows with java drivers in the past, I can already see great improvement moving it to Linux.
Well, my ignorance on software is a fact to begin with... ARTA/windows make me feel more confident, I can find my way with that. With REW I was forced to move to Linux after struggling with Windows. I will check again with wasapi exclusive. Regarding Linux, indeed there's no need to install any external drivers. And in REW preferences, I choose "DIYINHK". Is there anything more I should do?
Thanks! I decided for BNC because they do not occupy much space and they seem to be universal choice for bench instruments. So, I can use many interconnects, even scope probes. That said, I did build a set of custom interconnects with twisted screened cable -the same found in XLR. At the BNC end, only the twisted cores are connected. At the other end there will be an RCA or XLR or TRS wired accordingly. My conclusion is that it is impossible to create ground loop noise other than that coming from the DUT.

When the soundcard is to be used for low signal level/low distortion measurements, locking connectors (BNC, Lemo) ensure reliability/repeatability.
LEMOs are very expensive. BNCs are affordable
Unlocking TRS connectors can (and do) cause distortion indication fluctuations (contact issues).
What about 0.001% instead of 0.0001% ?
😉
George
Yes. Very unreliable contact in case we need to measure very low distortion level. XLR is hundred times better.Unlocking TRS connectors can (and do) cause distortion indication fluctuations (contact issues).
- Home
- Design & Build
- Equipment & Tools
- DIY soundcard intended for measuring amplifiers