* Run the program with main.py

For the first example we will showcase how you can use the full features of the package with main.py. Simply executing the main.py by giving the name of .csv file that contains the positional data of the satellite, as an argument in the function process(data_file):

def process(data_file, error_apriori):
    Given a .csv data file in the format of (time, x, y, z) applies both filters, generates a filtered.csv data
    file, prints out the final keplerian elements computed from both Lamberts and Interpolation and finally plots
    the initial, filtered data set and the final orbit.

        data_file (string): The name of the .csv file containing the positional data
        error_apriori (float): apriori estimation of the measurements error in km

        Runs the whole process of the program

Simply input the name of the .csv file in the format of (time, x, y, z) and tab delimiter like the orbit.csv that is located in the src folder and the process will run. You also need to input a apriori estimation of the measurements errors, which in the example case is 20km per point (points every 1 second). In the case you are using your own positional data set you need to estimate this value and input it because it is critical for the filtering process:

run = process("orbit.csv")


If the format of you data is (time, azimuth, elevation, distance) you can use the input_transf function first and be sure that the delimiter for the data file is tab delimiter since this is the one read_data supports.

The process that will run with the use of the process function is, first the program reads your data from the .csv file then, applies both filters (Triple moving average and Savintzky - Golay), generates a .csv file called filtered, that included the filtered data set, computes the keplerian elements of the orbit with both methods (Lamberts - Kalman and Spline Interpolation) and finally prints and plots some results. More specifically, the results printed by this process will be first the sum and mean value of the residuals (difference between filtered and initial data), the computed keplerian elements in format of (a - semi major axis, e - eccentricity, i - inclination, ω - argument of perigee, Ω - right ascension of the ascending node, v - true anomaly) and a 3d matplotlib graph that plots the initial, filtered data set and the final computed orbit described by the keplerian elements (via the interpolation method).


  • Reads the data
  • Uses both filters on them (Triple moving average and Savintzky - Golay )
  • Generates a .csv file called filtered that includes the filtered data set
  • Computes keplerian elements with both methods (Lamberts - Kalman and Spline Interpolation)
  • Prints results and plot a 3d matplotlib graph


  • Sum and mean of the residuals (differences between filtered and initial data set)
  • Final keplerian elements from both methods (first column : Lamberts - Kalman, second column : Spline Interpolation)
  • 3d matplotlib graph with the initial, filtered data set and the final orbit described by the keplerian elements from Spline Interpolation


Measurement unit for distance is kilometer and for angle degrees

The output should look like the following image.


* Run the program with automated.py

automated.py is another flavour of main.py that is supposed to run on a server. It keeps listening for new files in a particular directory and processes them when they arrive.


All the processing invloved in this module is identical to that of main.py.

For testing purpose some files have already put in a folder named src. These are raw unprocessed files. There is another folder named dst which contains processed files along with a graph saved in the form of svg.

To execute this script, change the directory to the script’s directory:

cd orbitdeterminator/

and run the code using python3:

python3 automated.py

and thats it. This will keep listening for new files and process them as they arrive.



  • Initialize an empty git repository in src folder
  • Read the untracked files of that folder and put them in a list
  • Process the files in this list and save the results(processed data and graph) to dst folder
  • Stage the processed file in the src folder in order to avoid processing the same files multiple times.
  • Check for any untracked files in src and apply steps 2-4 again.

* Using certain modules

In this example we are not going to use the main.py, but some of the main modules provided. First of all lets clear the path we are going to follow which is fairly straightforward. Note that we are going to use the same orbit.csv that is located inside the src folder and has tab delimeter (read_data.py reads with this delimiter).


  • Read the data
  • Filter the data
  • Compute keplerian elements for the final orbit

So first we read the data using the util/read_data.load_data function. Just input the .csv file name into the function and it will create a numpy array with the positional data ready to be processed:

data = read_data.load_data("orbit.csv")


If the format of you data is (time, azimuth, elevation, distance) you can use the util/input_transf.spher_to_cart function first. And it is critical for the x, y, z to be in kilometers.

We continue by applying the Triple moving average filter:

data_after_filter = triple_moving_average.generate_filtered_data(data, 3)

We suggest using 3 as the window size of the filter. Came to this conclusion after a lot of testing. Next we apply the second filter to the data set which will be of a larger window size so that we can smooth the data set in a larger scale. The optimal window size for the Savintzky - Golay filter is being computed by the function golay_window.c(error_apriori) in which we only have to input the apriori error estimation for the initial data set (or the measurements error):

error_apriori = 20.0
c = golay_window.c(error_apriori)

window = len(data) / c
window = int(window)

The other 2 lines after the use of the golay_window.c(error_apriori) are needed to compute the window size for the Savintzky - Golay filter and again for the polynomial parameter of the filter we suggest using 3:

data_after_filter = sav_golay.golay(data_after_filter, window, 3)

At this point we have the filtered positional data set ready to be inputed into the Lamberts - Kalman and Spline interpolation algorithms so that the final keplerian elements can be computed:

kep_lamb = lamberts_kalman.create_kep(data_after_filter)
kep_final_lamb = lamberts_kalman.kalman(kep_lamb, 0.01 ** 2)
kep_inter = interpolation.main(data_after_filter)
kep_final_inter = lamberts_kalman.kalman(kep_inter, 0.01 ** 2)

With the above 4 lines of code the final set of 6 keplerian elements is computed by the two methods. The output format is (semi major axis (a), eccentricity (e), inclination (i), argument of perigee (ω), right ascension of the ascending node (Ω), true anomaly (v)). So finally, in the variables kep_final_lamb and kep_final_inter a numpy array 1x6 has the final computed keplerian elements.


If the orbit you want to compute is polar (i = 90) then we suggest you to use only the interpolation method.

Using ellipse_fit method

If a lot of points are available spread over the entire orbit, then the ellipse fit method can be used for orbit determination. The module kep_determination.ellipse_fit has two methods - determine_kep and plot_kep. As the name suggests, determine_kep is used to determine the orbit and plot_kep is used to plot it. Call determine_kep with:

kep,res = determine_kep(data)

where data is a nx3 numpy array. The ellipse_fit method does not use time information at all. Hence, the input format is [(x,y,z),…]. The method results two arguments - the first output is the Keplerian elements while the second output is the list of residuals.

Plot the results using the plot_kep method. Call it with:


where kep is the Keplerian elements we got in the last step and data is the original data. The result should look like this.


Using propagation modules

Cowell Method

The module propagation.cowell propagates a satellite along its orbit using numerical integration. It takes into account the oblateness of the Earth and atmospheric drag. The module has many methods for calculating drag and J2 acceleration, and integrating them. However, here we will discuss only the important ones. One is propagate_state and the other is time_period. propagate_state propagates a state vector from t1 to t2. time_period finds out the nodal time period of an orbit, given a state vector. Call propagate_state like this.:

sf = propagate_state(si,t0,tf)

where si is the state at t0 and sf is the state at tf.


In all propagation related discussions a state vector is the numpy array [rx,ry,rz,vx,vy,vz].

Similarly to find out time period call time_period like this.:

t = time_period(s)

DGSN Simulator

The module propagation.dgsn_simulator can be used for simulating the DGSN. Given a satellite, it propagates the satellite along its orbit and periodically outputs its location. The location will have some associated with it. Observations will also not be exactly periodic. There will be slight variations. And sometimes observations might not be available (for example, the satellite is out of range of the DGSN).

To use this simulator, 3 classes are used.

  • The SimParams class - This is a collection of all the simulation parameters.
  • The OpWriter class - This class tells the simulator what to do with the output.
  • The DGSNSimulator class - This is the actual simulator class.

To start, we must choose an OpWriter class. This will tell the simulator what to do with the output. To use it, extend the class and override its write method. Several sample classes have been provided. For this example we will use the default print_r class. This just prints the output.

Now create a SimParams object. For now, only set the kep, epoch and t0.:

epoch = 1531152114
t0 = epoch
iss_kep = np.array([6785.6420,0.0003456,51.6418,290.0933,266.6543,212.4306])

params = SimParams()
params.kep = iss_kep
params.epoch = epoch
params.t0 = t0

Now initialize the simulator with these parameters and start it.:

s = DGSNSimulator(params)

The program should start printing the time and the corresponding satellite coordinates on the terminal.


The module propagation.simulator is similar to this module. The only difference is that it doesn’t add any noise. So it can be used for comparison purposes.

Kalman Filter

The module propagation.kalman_filter can be used to combine observation data and simulation data with a Kalman Filter. This module keeps on checking a file for new observation data and applies the filter accordingly. We can use the DGSN Simulator module to create observation data in real time. First, we must setup the simulator. We must configure it to save the output to a file instead of printing it. For this, we will use the in-built save_r class.

Run the simulator with the following commands.:

epoch = 1531152114
t0 = epoch
iss_kep = np.array([6785.6420,0.0003456,51.6418,290.0933,266.6543,212.4306])

params = SimParams()
params.kep = iss_kep
params.epoch = epoch
params.t0 = t0
params.r_jit = 15
params.op_writer = save_r('ISS_DGSN.csv')

s = DGSNSimulator(params)

Now the program will start writing observations into the file ISS_DGSN.csv. Now we need to setup the Kalman Filter with the same parameters. Use util.new_tle_kep_state to convert Keplerian elements into a state vector. In this tutorial, it is already done. Run the filter by passing the state and the name of the file to read.:

s = np.array([2.87327861e+03,5.22872234e+03,3.23884457e+03,-3.49536799e+00,4.87267295e+00,-4.76846910e+00])
t0 = 1531152114

The program should start printing filtered values on the terminal.

Using utility modules


new_tle_kep_state is used to convert a TLE or a set of Keplerian elements into a state vector. To convert a TLE make an array out of the 2nd line of the TLE. The array should be of the form:

  • tle[0] = inclination (in degrees)
  • tle[1] = right ascension of ascending node (in degrees)
  • tle[2] = eccentricity
  • tle[3] = argument of perigee (in degrees)
  • tle[4] = mean anomaly (in degrees)
  • tle[5] = mean motion (in revs per day)

Now call tle_to_state. For example:

tle = np.array([51.6418, 266.6543, 0.0003456, 290.0933, 212.4518, 15.54021918])
r = tle_to_state(tle)

Similarly a Keplerian set can also be converted into a state vector.


teme_to_ecef is used to convert coordinates from TEME frame (inertial frame) to ECEF frame (rotating Earth fixed frame). The module accepts a list of coordinates of the form [t1,x,y,z] and outputs a list of latitudes, longitudes and altitudes in Earth fixed frame. These coordinates can be directly plotted on a map.

For example:

ecef_coords = conv_to_ecef(np.array([[1521562500,768.281,5835.68,2438.076],

The resulting latitudes and longitudes can be directly plotted on an Earth map to visualize the satellite location with respect to the Earth.