J
Johannes Nix
Hi,
this might be of interest for people who are look for practical
information on
doing real-time signal processing, possibly using multiple CPUs, and
wonder
whether it's possible to use Python for audio-type worst case
latencies (around 25 ms).
I've done that in my PhD work, both with real-time requirements on
dual-CPU
64 bit platforms, and with very complex algorithms running on
multicomputers.
What I found is that numerical Python is a great environment for such
tasks. I've used it as well for massively parallel algorithms
(particle filters) for simulations of auditory scene analysis. What is
a very
special advantage is that if you get faster hardware, you can simply
copy your algorithms to a new system and compile - even if
it has a different CPU!
I've documented the approach in my PhD thesis, in Appendix A,
starting with some thoughts on developments in signal processing
in the last years. This piece is available online. Title and abstract
of that
chapter read as follows:
--------------------------------------------------------------
A real-time, script-based, multiprocessing Solution for experimental
Development of Signal Processing Algorithms
Evaluation of audio signal processing algorithms on real-time
platforms has
unique advantages. However, such environments also used to have the
disadvantage
of requiring expensive hardware, and tedious work to set them up,
while providing only a short useful life. This report proposes to
exploit advances
in hardware and software development by integrating real-time
processing
with script-based explorative development and use of multiprocessing
hardware.
The concept was implemented based on standard hardware and open
source software, and its realization and characteristics are presented
here. Applications
of the system for algorithm development and evaluation are described
briefly.
--------------------------------------------------------------
Here is the download link for several paper formats:
http://medi.uni-oldenburg.de/members/jnix/index.html#thesisdownload
Alternatively, for ISO A4 paper, use one of these two URLs:
http://medi.uni-oldenburg.de/download/paper/Nix,Johannes-PhDthesis-2005-ISO-A4-format.pdf
http://docserver.bis.uni-oldenburg.de/publikationen/dissertation/2006/nixloc05/nixloc05.html
(for that paper size, this are the PDF pages 155 - 163)
If you want to cite the chapter, e.g. when doing advocacy for
scientific
computing using SciPy, please do this as follows:
Nix, Johannes (2005), "A real-time, script-based, multiprocessing
Solution for experimental
Development of Signal Processing Algorithms", in: Localization and
Separation of Concurrent Talkers Based on Principles of Auditory Scene
Analysis and Multi-Dimensional Statistical Methods, Appendix A, Ph.D.
thesis, Universität Oldenburg, Germany.
Also, I am currently looking for interesting further work
opportunities
or contracts in the domain of scientific computing and statistical
estimation.
If you know some interesting position, don't hesistate to contact me.
Kind regards,
Johannes
--
Dr. Johannes Nix
Energy & Meteo Systems GmbH
Research & Development of windpower forecasts
Bremen, Germany
Phone: + 49 421 8963914
this might be of interest for people who are look for practical
information on
doing real-time signal processing, possibly using multiple CPUs, and
wonder
whether it's possible to use Python for audio-type worst case
latencies (around 25 ms).
I've done that in my PhD work, both with real-time requirements on
dual-CPU
64 bit platforms, and with very complex algorithms running on
multicomputers.
What I found is that numerical Python is a great environment for such
tasks. I've used it as well for massively parallel algorithms
(particle filters) for simulations of auditory scene analysis. What is
a very
special advantage is that if you get faster hardware, you can simply
copy your algorithms to a new system and compile - even if
it has a different CPU!
I've documented the approach in my PhD thesis, in Appendix A,
starting with some thoughts on developments in signal processing
in the last years. This piece is available online. Title and abstract
of that
chapter read as follows:
--------------------------------------------------------------
A real-time, script-based, multiprocessing Solution for experimental
Development of Signal Processing Algorithms
Evaluation of audio signal processing algorithms on real-time
platforms has
unique advantages. However, such environments also used to have the
disadvantage
of requiring expensive hardware, and tedious work to set them up,
while providing only a short useful life. This report proposes to
exploit advances
in hardware and software development by integrating real-time
processing
with script-based explorative development and use of multiprocessing
hardware.
The concept was implemented based on standard hardware and open
source software, and its realization and characteristics are presented
here. Applications
of the system for algorithm development and evaluation are described
briefly.
--------------------------------------------------------------
Here is the download link for several paper formats:
http://medi.uni-oldenburg.de/members/jnix/index.html#thesisdownload
Alternatively, for ISO A4 paper, use one of these two URLs:
http://medi.uni-oldenburg.de/download/paper/Nix,Johannes-PhDthesis-2005-ISO-A4-format.pdf
http://docserver.bis.uni-oldenburg.de/publikationen/dissertation/2006/nixloc05/nixloc05.html
(for that paper size, this are the PDF pages 155 - 163)
If you want to cite the chapter, e.g. when doing advocacy for
scientific
computing using SciPy, please do this as follows:
Nix, Johannes (2005), "A real-time, script-based, multiprocessing
Solution for experimental
Development of Signal Processing Algorithms", in: Localization and
Separation of Concurrent Talkers Based on Principles of Auditory Scene
Analysis and Multi-Dimensional Statistical Methods, Appendix A, Ph.D.
thesis, Universität Oldenburg, Germany.
Also, I am currently looking for interesting further work
opportunities
or contracts in the domain of scientific computing and statistical
estimation.
If you know some interesting position, don't hesistate to contact me.
Kind regards,
Johannes
--
Dr. Johannes Nix
Energy & Meteo Systems GmbH
Research & Development of windpower forecasts
Bremen, Germany
Phone: + 49 421 8963914