Skip to content
Audio measurement workstation
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
controller Initial commit Mar 21, 2018
designs Initial commit Mar 21, 2018
images IMD image Mar 24, 2018
preamp update schematics Oct 4, 2018
.gitignore Initial commit Mar 21, 2018
LICENSE Create LICENSE Mar 23, 2018 Update Mar 24, 2018

Audio Measurement Workstation

Initial Specifications

  • Class II appliance - no ground loops incl. display
  • Balanced inputs for measuring bridged or digital amps
  • Dedicated device
  • Decent measurement quality - PCI soundcard ESI Juli@

Design Overview

Thin Client PC FS Futro S900N

  • Class II power adapter
  • PCI slot incl. +12/-12V supply voltages
  • Dual core CPU to allow measurement and VNC remote access
  • Connectors for two SATA drives to allow increasing reliability/repairability with RAID1 configuration
  • No monitor, keyboard, mouse - access via network VNC
  • Current IP address relayed via speech synthesis announcements, using internal soundcard and speaker.
  • Inexpensive from ebay

ESI Juli@

  • Both balanced and single-ended inputs and outputs, functioning simultaneously. It allows feeding an amp with single-ended signal and measure balanced speaker output at the same time.
  • Digital outputs/inputs for extra measurements/tests
  • Full support in linux
  • Priced moderately on ebay due to lack of PCI slots in modern PCs


  • Linux Mint Mate (v. 18 at the time of writing)
  • Arta measurement software running under wine
  • Control - python script hooked to ACPI power button.
    • At startup - Etherner and/or wifi IP addresses are announced.
    • Pushing the power button once - IP addresses announced
    • Pushing the power button twice within two seconds - shutting down + announcement


  • Measurement mode - remotely via VNC on another PC
  • Admin mode - possible to connect monitor/keyboard/mouse at any time


All parts are 3D printed, FreeCAD sources and STL models are included in design folder.

Since the thin client supports only half-size PCI cards, room/holes must be made for full-sized one. Upper sound jacks and PS/2 slot must be removed/sawn off.

The original holder of the secondary heat sink must be removed, the heat sink fixed directly to the bottom with a new holder.

The soundcard is connected with a PCI cable adapter ( It is mounted with 3D-printed brackets. The lower brackets provide only shielding, using copper tape connected to the PC enclosure. Coax digital input and output are extracted from the multiconnector socket to individual add-on cinches.

The back panel is printed in two colors.

Holes for long cinches are drilled into the front panel, covered with a two-color printed plate.


Single-ended and balanced inputs are mixed before the ADC. Therefore, for best performance of the higher-voltage balanced inputs short-circuit the more sensitive single-ended inputs with short-circuited cinches. Inexpensive at . In balanced-balanced closed-loop measurement this improves noise RMS from -102dB to -105dBFS.

Closed loop no signal

Closed loop 1kHz signal

Closed loop two sine 13+14kHz signal

Amps are measured with a dummy load with replaceable balanced voltage dividers made from old CPU heatsinks, optionally cooled with the original heatsink fans. I will post more pictures and resistor values.


Testing Juli@ Voltage Supplies

The PC generates +12V/-12V voltages for the PCI slot. I cut these supply pins on the connector between digital and ADC/DAC parts of Juli and fed the ADC/DAC part with 7812/7912 linear PSUs. No change in noise background was detected in Arta. It means the internal regulators on Juli are capable of filtering the SMPS-generated supply voltage sufficiently and the PCI-supplied voltages can be used safely.

Shielding Juli@

I measured the closed-loop noise floor before covering the bottom holders of Juli@ with copper tape and connecting to the case. The difference between the unshielded and shielded soundcard was less than 0.2dBFS. Therefore I assume the bottom shields are not necessary and do not have to be printed/installed at all.

No Impact of SSD on Measurement Performance

Running full RAID1 sync on the secondary SSD located right beneath the soundcard had no measurable impact on the closed-loop noise floor of 105.1dBFS.

VNC Resolution

With no monitor connected the X server starts at 1024x768. This resolution seems just enough for taking measurements/pictures while keeping the CPU load for VNC transmission acceptable.

You can’t perform that action at this time.