Vector Implementation of Multiple TLEs

[cassova]

I'm looking for a way to pass in an array (like numpy multi-dim array) that contains multiple TLEs for same or diff satellites. The goal is to take advantage of vector performance to get an e array (satellite position) based on a separate array of jd,fr (one for each TLE).

It looks like SatrecArray only takes in instances of already initialized satellites. Can this be called without having to create Satrec for each satellite?

Thanks,
Nicholas

[brandon-rhodes]

I am not aware of any way except to create a Satrec for each satellite ­— that is the only class which is used to represent a satellite; there is not an alternative.

Are you finding that the expense of creating one Satrec for each satellite is too slow, or is using too much memory?

[cassova]

It might work for me as is. For now, this is what I'm doing using a pandas dataframe:

from sgp4.api import Satrec
from sgp4.conveniences import jday_datetime
from datetime import datetime, timezone

# Create the satellite object (used to find satellite position)
dfs['satobj'] = dfs.apply(lambda x: Satrec.twoline2rv(x['TLE_LINE1'], x['TLE_LINE2']), axis=1)

# Get the Julian date of the EPOCH
dfs.reset_index(inplace=True)
dfs['epoch_julian'] = dfs['EPOCH'].apply(lambda x: jday_datetime(x.replace(tzinfo=timezone.utc)))

# Get the days since 1949 December 31 00:00 UT
# This will be used when creating satobj for the test set
# (this is needed to get the satellite position from generate TLEs)
ref_date = datetime.strptime('12/31/1949 00:00:00', '%m/%d/%Y %H:%M:%S')
dfs['epoch_days'] = dfs['EPOCH'].apply(lambda x: (x-ref_date)/np.timedelta64(1, 'D'))
dfs.set_index('EPOCH', inplace=True)

# Get satellite x,y,z positions
dfs['satpos'] = dfs.apply(lambda x: x['satobj'].sgp4(*x['epoch_julian'])[1], axis=1)

On a sample set of 241687 unique TLE records, this takes about 14s on my machine. While this is reasonable, my end goal is to do this on all TLE data (like gp_history all). I can probably wait a few hours if that's what it's going to take.

[cassova]

Checking the memory, it looks like that could be an issue.

TLE_LINE1               30452562
TLE_LINE2               30452562
satobj                 245553992
epoch_julian            11600976
epoch_days               1933496
satpos                  13534472

[brandon-rhodes]

I will be interested to hear how performance changes (if at all) once you have the chance to try out the SatrecArray. I'm not sure, alas, that there's any way around the memory usage, as down at the C++ level those satellite structs just have lots of floats stuffed into them.

[cassova]

Unfortunately SatrecArray applies the same jd,fr dates to all satellites in the array. While I need to apply a unique value to each so I'm not able to use it.

[brandon-rhodes]

Happily, the opposite is the case: it applies a different jd and fr to each satellite. Could you suggest how the docs at the following link might be improved to make that clearer? I'll be happy to revise them:

https://pypi.org/project/sgp4/#array-acceleration