Please click on the images for the full versions.

This wiki presents my current best knowledge of this issue. It can not be guaranteed that all conclusions are correct. This also means that the identification of the flight and aircraft may be wrong.

In the analysis of data set L2011_23410, I found that some strong RFI spikes were not flagged in my initial setup of the flagging routine. It turned out that these integrations had consistent signals with rather small delays. The initial thought was that we see some strong transient in the sky, e.g. giant pulses from the crab pulsar. For the further analysis I switched off the flagging completely to reveal many more similar spikes, with a large range of delays.

This plot illustrates the distribution in time and frequency in the autocorrelations of station RS106HBA:

Versions at higher resolution are available as postscript and PDF. The gaps are due to a temporary technical problem with half of the subbands. We have alternating blocks of 30 good and 31 bad subbands.

Note that the spikes are very wide-band, extending up to ca. 141 MHz. This is outside of the airband, and communication signals are generally narrow-band.

There are no similar spikes outside of this time range, but many within cannot be seen in this visualisation.

The spikes do not affect adjacent integrations, which means that they must be (much) shorter than 1sec.

I tried to determine the position by mapping delay spectra of several baselines to the sky, but without finding a clear peak. The spikes are certainly not coming from the pulsar.

In the next attempt I tried to convert the delays to positions on the ground. This time there were areas with a stronger signal, but no clear compact peak. Because I found that the position of “blobs” in these maps were moving with time, I thought about aircraft signals and imaged the area around LOFAR at a height of 10200m. Here are two of these maps, separated by 112sec (time labels after start of observations at 14:00 UTC):

The origin for X and Y is 6.86 deg longitude and 52.9 latitude. To produce these images, data from all Dutch baselines were first converted to delay-spectra (signal strength as function of delay). For a range of positions (X,Y,Z), geometric delays were calculated for all baselines and the signals were averaged incoherently (because of unknown phase errors). Weights for the individual baselines were set to the reciprocal maximum values to account for different sensitivities and frequency-averaging amplitude losses. The curves seen in the images are hyperbolas that correspond to measured delays on different baselines. The total signal is strongest where they intersect.

Only the lowest 8 subbands (1.6 MHz) of the second block were used for these two plots.

Because of projection effects, the altitude can only be determined accurately at small distances. I scanned a range of X,Y,Z (with 12.5m steps) for a limited time range (21.75-21.8h) using the second block of good 30 subbands (6 MHz). The best positions of those time steps with RFI spikes are plotted here:

The speed in this time range is about 780 km/h.

Here is a google map with overlayed LOFAR stations and positions where the aircraft was detected near LOFAR. The airplane's trajectory is marked with red pins, the used LOFAR stations are green. Moving the mouse pointer over the aircraft pins shows the time.

Because fitting for X,Y and Z is expensive, the altitude was kept constant at 10200m for the following fits. This is supposed to be a reasonable approximation. I scanned X/Y within +-300 km (stepsize 100m) using the lowest 8 subbands of the second block for each integration in the time range 21:15-22:00 (no signal outside of this range). Here I plot the positions only of those peaks with a amplitude well above the general noise level:

Older versions of these plots can be found here and here. Those positions were determined with less resolution and are not that accurate.

You may want to compare the station positions with this map:

Here is a google map with overlayed LOFAR stations and positions where the aircraft was detected. The airplane's trajectory is marked with red pins, the used LOFAR stations are green. Moving the mouse pointer over the aircraft pins shows the time. An old version of that map can be found here.

The average speed of the aircraft is 800 km/h. Note that the trajectory is not exactly straight. In particular do we see a turn at about 21.8 h when the aircraft just passes over central LOFAR. At that time there are no signals for a bit more than one minute.

At larger distances the accuracy is quite low. The wiggles and jumps are probably not real. With more subbands it would be possible to increase the accuracy.

The first detection (not quite included in these plots) is at a distance of about 350km, which is pretty close to the horizon at 10km height.

The peak in XYZ space is not so easy to visualise. Here I use 2-dimensional cuts through the peak for the time 28017sec (of the observation) determined for 30 subbands (second block):

These results were presented at the LOFAR Status meeting on 23 March 2011.

As pointed out by Hans van der Marel, the data seem to agree with those of Aeroflot flight AFL231 from Moscow to Brussels. The data of that (and other) flights can be checked here. They seem to agree to within a few hundred metres and several seconds. I investigated this a bit further and found here that the aircraft itself has the registration VP-BUO. Thanks to millions of plane spotters, there are many pictures of this aircraft on the web. The following was taken from here:

— Olaf Wucknitz 29-Mar-2011 11:06

I also checked the data sets L2011_22883 and L2011_22899 for simular signals. In the latter I did not find anything like this, but L2011_22883 (3C196 on 13/14 Jan 2011 21:30-03:30, using only every 4th subband, with total number of 60) showed some wide-band pulses as well, even though they are not as obvious as the ones discussed above.

Between 19:01:30 and 19:11:24 UTC I find 7 spikes that seem to be consistent with another aircraft trajectory:

For this analysis I combined the subbands incoherently, because the coherent “beam” has many local maxima of similar level as result of the gaps in frequency which can lead to false detections. This reduces the accuracy.

The mean altitude is 8800 m. A linear fit results in a vertical speed of 4.3 m/s, but this may not be significant. The mean horizontal speed is 690 km/h.

Here is a google map with overlayed LOFAR stations and positions where the aircraft was detected. This time I cannot find an aircraft with similar trajectory, but the data base is known to be incomplete.

The spectral shape of the signals seems to be different from the earlier ones, but they again spread over several MHz.

— Olaf Wucknitz 01-Apr-2011 12:47

Axel Jessner (MPIfR) reports that both Aeroflot and Eurocontrol are aware of the problem, as shown in a report titled Status of AEROFLOT Squelch Burst Interference Issue, issued by Eurocontrol. The equipment that produces the interference is identified, but a fix seems to be difficult.

There is an ASTRON report written by Hans van der Marel and myself that summarises the findings. See also ASTRON/JIVE daily image 13 May 2011.

— Olaf Wucknitz 26-May-2011 13:23