Thursday, November 28, 2013

FM passive radars as alternatives to Graves

Two weeks ago, on November 13, I left my computer logging meteors on Graves' frequency with the help of Baudmeteors. My goal was to catch the Leonids shower. But on November 15, around 16:40 UTC I detected one meteor and then, only silence for the next two days. Graves was turned off.

You don't realize how dependent the radio meteor detection is from Graves until you lost it. Around 20 UTC I realized there was no problem with my radio or just a very low meteor count. Graves were really off!

What are the alternatives to Graves? Sadly for us, there are very few.

A good transmitter for radio meteor detection must run continuously over time and be high power. And, if you are interested in Doppler measurements, it must have a stable carrier somewhere. The best operating frequency is near the 50 MHz band, so the further away from 50 MHz, less attractive the transmitter is. Band I TV transmitters were perfect for this kind of job, but now there is not a single Band I TV transmitter in western Europe. With TV outside the 50 MHz area, most countries allowed the use of the 50 MHz band for hams, and there are many amateur beacons in this band now, most of them between 50.0-50.1 and 50.4-50.5 MHz, but they seems to run too low power to be useful for radio meteor detection. I was never be able to receive any of these beacons via meteors.

Another possibility are aeronautical navigational VOR transmitters, between 108 and 118 MHz. They have a nice carrier, so Doppler measurement is easy, but they are "low" power. The high power VOR transmitters runs about 200W EIRP. I have been able to detect meteors using VOR signals, but they are somewhat disappointing, specially after having meet Graves: With Graves you get a meteor every few seconds but with VOR transmitters I get one or two meteors in a whole night.

Another option is the use of FM broadcast signals in the 87.5-108 MHz range. They are powerful, but lacks a defined carrier, so Doppler measurements can't be made with them. Another problem is near large cities, this band is so crowded that there is virtually not a single clear spot where look for distant signals.

But recently a new approach has appear.

FM broadcast passive radars

The DVB-T dongles based on the RTL2832 chip have revolutionised the ham world in many ways. For about $10, shipping included, you get a simple but effective software defined radio (SDR) capable of doing things really far beyond the analogical receivers. Most people use SDR receivers in the same way they use a classic receiver, which is a waste of resources. The power of a SDR receiver resides in the ability to process the signals. And processing the signals very interesting things are possible.

In September 2013 Juha Vierinen posted the results of an experiment he made with a pair of DVB-T dongles sharing a common clock and the signal from a FM broadcast station. He was able to detect airplanes, but he was also able to detect meteors. Is this a new form to detect meteors using radio waves?

The working principle is quite simple. One receiver with its antenna points to the transmitter and gets the direct signal, and the other receiver, with its own and separated antenna, points to another area where you expect to find the echoes.

Then, the processing software looks for the direct signal in the echo signal. Of course it will find it. Real world antennas are not like laser beams, and both antennas will receive the direct signal. But the processing software also looks for delayed copies of the direct signal in the echo receiver, as well for Doppler shifted signals.

For example, if you have a signal reflection from an airplane, this signal will arrive to your receiver some time later than the direct signal, because its traveled distance is also greater. Also, if the airplane is moving ( if it is flying, it is moving ), that reflected signal will have a Doppler shift.

The key to process these signals is cross-correlation. The software looks for a combination of delayed and Doppler shifted copies of the direct signal in the echoes signal and fills a matrix, who later is converted to an image. In one axis you have the Doppler shift ( = bistatic target speed ), and in the other axis you have the delay ( = bistatic range ). This is the reason at doppler = 0, distance = 0 there is a peak. This peak is the direct signal received in the echoes antenna.

I contacted with Juha and he kindly provided me a set of samples files to play with them. I processed the files using Octave and after some days of trial and error I got this:

If you look carefully you can see some interesting things. There are two moving targets. One at about 20Hz/120km, and another one moving fast from about 25Hz to almost 0 around 30km. No doubt they are airplanes.

There are also some static ( at 0Hz ) reflections. One quite weak around 120km and another one, much stronger around 180km. Juha confirmed me they are reflections from mountains. I was very happy because I implemented the algorithm from scratch and it seems to work.

But know the problem arises. Juha used two receivers with two antennas. I only have one antenna at the roof, so I could not replicate Juha's experiment.

Single antenna FM broadcast passive radars

Some days later, I tried to cross-correlate a signal with itself. This is the directed signal with itself and the echoes signal with itself. This is the result of cross-correlate Juha's direct signal with itself:

Nothing interesting. And then, I repeated the process using only Juha's echoes signal:
Surprise! The planes were there and the reflections from mountains are there too. How is this possible?

It is possible because operating with a single channel, the process is just to look for echoes of the same signal instead of looking for a signal into another signal. At the end, it is almost exactly the same, but this allow me to replicate Juha's experiment with my antenna.

After many and many hours of tests, with different transmitters, I had no luck. Despite I live very near Madrid and dozens of airplanes are flying above my head every second, I was not able to detect any of them. The only success I had was some static reflections using some FM signals, but nothing more.

What was the problem?

I suspect there was not a single problem, but some of them. The first problem is the strong signal from the FM transmitters coming from my antenna.

The FM broadcast band at the 144MHz output of my MX2000 triplexer

You can see the band pass effect of my triplexer, attenuating the signals at 88MHz but letting pass 108MHz signals without too much problem. In fact the problem is there are some signals over -30dBm. With all these big signals, the small tuner in the DVB-T dongle will suffer for strong intermodulations and sure from image responses: the R820T tuner uses a 4 MHz IF. All this will reduce the chances to detect the weak signals reflected from airplanes or meteors.

The other problem is related to the design of the RTL2832 chip. It uses a 8 bit ADC, so the dynamic range of this chip is just a bit less than 50 dB. This means if the echo signal is more than 50 dB weaker than the direct signal, RTL2832's ADC will not see it. And this is assuming the tuner and RTL AGC regulates the signal exactly to have the full ADC range: From 0x00 to 0xFF. In real life, the AGC will regulate to a lower gain to prevent clipping, reducing further the dynamic range, and making the detection of echoes more difficult.

At this point one question arises: Why Juha was successful?

Watching carefully Juha's videos in YouTube, I suspect Juha did the experiment in a relative quiet RF area, and used a relatively distant FM signal, maybe with obstructed direct path. so the direct signal was comparable to the echoes signals in strength.

If I am right, maybe this technique is useful for amateurs living in rural areas, with low density of FM broadcast transmitters around, who can receive a weak direct signal from a remote transmitter in a clear frequency. For amateurs like me, living near big cities with a jungle of really strong of FM signals, a DVB-T dongle does not work.

I suspect I could replicate Juha's experiment using a rock-solid front end, and a 16 or 24 bit SDR receiver. Maybe a common FM receiver tapping into its 10.7 MHz IF with a good quality 16 or 24 bit HF SDR will do the work. Unfortunately now I only have DVB-T dongles now.

Another item in the huge TO-DO list.


Now, Graves is the only signal in Europe feasible to serious radio meteor work. If Graves is turned off during a big shower, or if Graves is closed down and disappears forever, there is no other reliable signal.

We need to develop another way to detect meteors via radio, and the FM broadcast signals are the higher power common transmitters now in VHF. Let's use them, even if we need digital processing techniques.

By the way. Graves returned to normal operation around 10 UTC on November 18.


  1. Any chance you could share the exact details of your hardware and software setup? I've been searching for software to do this. I have some very good signal sources here but I haven't been able to resolve the software side of this yet.

    Best 73, and thanks for sharing your results here! Vy FB!

    Tim de NW0W

    1. Currently my code is very dirty and some tasks must be done by hand. It just analyze a file recorded with rtl_fm working in raw mode, and a Octave script to compute the individual frames. Then the animated gif is constructed using ImageMagick. I will publish it after a bit of cleaning, so stay tuned.

  2. Look at what we have done at RETRAM with two channels ( , sorry mainly in french).
    you can remove greatly increase the two channel system by adaptative filtering. The problem lies in the fact you cannot have completely two independant channels, you need to remove the 'common part'.
    we have done this successfully, works great (but signal processing is complex).
    Google "passive radar clutter removal" to have the processing details, or contact me

  3. Miguel, you are right, to repeat Juha case study we need strongly suppress direct path signal. 8 bit dynamic range is too small for the discrimination the week signal reflected from the target from strong direct signal from station.

    I did the same experiment with rtl-sdr dongle: record FM station I/Qs and analyse this stream in Octave with ambiguity function. I found other issue: ambiguity function sidelobes is still high even for 10 Hz Doppler resolution and we need to suppress direct path anyway.

    In Internet I met one technique named lattice prediction adaptive filter that allow to do that. In future, I am going to design receiver with 14-bit IF ADC to study passive radar technology.