On 03/06/2018 10:32 AM, Langston, Glen
wrote:
Dear Marcus,
Thanks for your efforts on spectra_radiometer.
I’m still pursuing the issues with alias spectra and
linux,
but I did see the same alias problem with
spectro_radiometer under linux.
I’ll download the latest version spectro_radiometer and
produce the mac and linux comparision
plots in the near future (maybe tonight).
FYI the effect I’m seeing with “NSF watch” aka (watch.py)
is shown by
comparison in the two attached screen captures. Odriod
shows the problem
and mac does not.
I’ve got to gather up the strength try to get my python
code into GitHub.
The current way I install is just tar -xvf the latest tar
file (attached),
cd into the directory and the type:
python watch.py
On Odroid I actually type
taskset -c 4,5,6,7 python watch.py
so as to use the fastest 4 of the 8 processors.
The odroid is able to eat all 6MHz bandwidth Airspy mini
data
The kernel will migrate processes as they start to
max out their respective CPU, so using "taskset" is generally not
required.
Note my major change/improvement over what happens
naturally
with the WxGUI FFT is to sum all of the FFTs that are
passed
as messages. The Gnuradio Wx Spectrum GUI only processes
the first of N messages in the packet. For radio
astronomy,
I’m averaging about 1.3 Million spectra per 4 minute
observation.
I appreciate the way you’ve put all the configuration
arguments on
the command line of spectro_radiometer.py. My
implementation
stores the all information needed to configure the entire
observation
in each observing file. (Note a data directory is
created
parallel to the “watch” directory).
My approach to having a "config file" functionality
is to wrap the "naked" .py in a shell script that can handle all
of that.
I’d appreciate any suggestions you, or anyone, has.
Best regards
Glen
Since some needed packages are not automatically installed
by gnu radio
(such as ephem, I’ve added “try” code with suggestions for
installing
the likely packages)
I have done that in the past, and I tend to get
sloppy about it. My odroid stuff I have total control over what's
on the system image,
so the corresponding GR flow-graphs don't need to care, because
they "know" that ephem is "just there."
The attached plots show the observation of RFI found at
1440 MHz in
my vicinity. The Linux gnu radio observations show the
problematic alias. The data files have
date and time names with extensions:
.ast - astronomy
.hot - hot load calibration observations, usually pointing
straight down (El = -90)
.not - configuration for observation setup. (Notes==.not)
Occasionally I compute a cold load spectrum for full
calibration:
.cld
Have you tried this on a Linux platform that isn't an
Odroid? I'm thinking of numerical bugs in things like VOLK or the
SIMD implementation or something like that.
But I haven't observed this imaging effect myself on any of my
Odroid+AirSpy systems. Our dual-channel H1 spectrometer is
AirSpyR2 + Odroid XU4 + Arch.
|