Midi Real-Time Performance Testing

Downloading our Performance Measurement Software:

Our Papers:

• M .Nelson and B. Thom, A Survey of Real-Time MIDI Performance, (ps,pdf).
• B. Thom and M. Nelson, An In-Depth Analysis of Real-Time MIDI Performance, (ps,pdf).

Related Papers:

• R. Becina and P. Burk, PortAudio--- an Open Source Cross Platform Audio API
• J. Wright and E. Brandt, System-Level MIDI Performance Testing
• J. Wright et al, Patent: 6546516, US Patent Office.
• A. Freed, A. Chaudhary, and B. Davila, Operating Systems Latency Measurement and Analysis for Sound Synthesis Processing Applications.
• E. Brandt and R. Dannenberg, Low-Latency Music Software Using Off-The-Shelf Operating Systems

Related Websites:

• Platform Independent Libraries for Sound and MIDI: PortMusic
• Open-source software and standards for creating music: OpenMuse

Building a Transcoder:

• Some pictures:
  • A photograph of our transcoder being used to test a 2x2 MidiSport
  • A photograph of our transcoder and the two proud parents
• A hand-drawn schematic of the transcoder we built. Important: Be sure to check the pin-outs in this schematic with the chips you buy! It was our experience that not all Opto Isolators that are sold as "having the same specs for MIDI" use the same pins.
• Parts list (we built ours for about $20):
  • (1) 5VDC Bread Board
  • (2) OPTO Isolators (6N138)
  • (1) Hex Inverter w/Open Collector Outputs (74LS05)
  • (1) Hex Inverter with Schmitt Trigger Inputs (74LS14)
  • (3) DIN 5-pin Male (MIDI Connectors)
  • (1) 1/4-inch Stereo Audio Jack.
  • (2) 270 Ohm Resistors
  • (6) 220 Ohm Resistors
  • (2) 0.047 microFarad Capacitors
• Recorded transcoder audio wave file
• Useful Related Resources:
  • The Midi Manufacturer's Association electrical spec
  • Midi Thru Box schematic

Details Concerning our Empirical Data:

Interfaces Tested:

• MOTU Fastlane (Motu), USB
• Miniman Midisport (2x2), USB
• EgoSys Miditerminal 4140 (4140), Parallel Port
• Creative Labs SoundBlaster Live! 5.1 (SBLive), PCI (MPU-401 compatible)

Operating Systems Tested:

• Linux with 2.4-series kernel (Linux 2.4) using Debian GNU/Linux distribution with ALSA 0.9.4, kernel 2.4.20, and low latency patches (Robert M. Lowe's variable-Hz (Hz=1000) and pre-emptible kernel patches and Andrew Morton's low-latency patch).
• Linux with 2.6-series kernel (Linux 2.6), as above but with ALSA 0.9.7, kernel 2.6.0, and no special patches.
• Mac OSX (OSX) 10.3.2 (Panther) with CoreMIDI.
• Windows 2000 (Win2k) SP4 with WinMME.
• Windows XP (WinXP) SP1 with WinMME.

Computers Tested:

• HP Pavilion 751n desktop (HP) with 1.8 Ghz Intel Pentium 4 processor and 256 MB RAM.
• Apple Mac G4 desktop (G4) with dual 500 MHz G4 processors and 320 MB RAM.
• IBM Thinkpad T23 laptop (T23) with 1.2 GHz Intel Pentium II processor and 512 MB RAM.

Interpreting Logs and Graphs

• A few words about Histogram Graphics: Blah, Blah, Blah.
• For more information about the test and ref log files, see the readme (where is this???)

TEST: Active Sense (35 ms periods, 1 hour duration)

TEST: Burst (100 ms periods, 10 note bursts, 1 hour duration)

TEST: Load Burst (100% CPU and Cache utilization, 100 ms periods, 10 note bursts, 1 hour duration)

Some Unfinished Exploration:

• Some temporal data collected from various rushed end-of-term experiments
• Some business about "worse" configurations on Windows not looking so bad.
• Some data for short vs. long vs. very long runs.