Over on YouTube user Albert Schäferle has uploaded a short video showing his reception of some radar pulses and their corresponding echoes. He uses rtl_fm and pipes the output into Baudline which is used to display the radar waveform. On the video description he writes:

Receiving direct and (supposedly) reflected pulses from an L-band radar in Učka, HR (Lockheed Martin AN/FPS-117). The receiving station was 83 km away, with clear LOS.
Center frequency is 1258 MHz (one out of four that this frequency-agile radar head is using).
The receiver is a RTL-SDR dongle (R820T tuner IC) with a 2-dipole collinear array (tuned for 403 MHz) and approx 7 m of Belden 1694A RG-6 coax.

rtl_fm output was piped to baudline, which is the software shown in the video. The IQ sampling rate is 2 MHz; the transform is a complex STFT (size=2048 samples, Blackman window).
This is a 0.008x speed playback of 15 ms of recording.
The (again, supposedly) reflected pulses are obviously more time-local with a shorter transform window size, e.g. 512 samples http://i.imgur.com/sAHWhwD.png

The effect of pulse compression is quite evident http://www.radartutorial.eu/08.transm…
The direct-reflected delay is approx 278 µs (~42 km from receiver, in a simple 2D, along beam, normal incidence model). I should add that this “reflection delay” effect does not usually show up.
There’s another fainter echo closer to the pulse, but I suspect that it could be a time-sidelobe of the main pulse: a side effect of pulse compression. Anyway, I must state that I have no formal knowledge on radar topics. So you’d better take all this with a grain of salt 😉

Link to recording: https://db.tt/Lxe67Ig3 (save destination as…)

Video recorded with VLC, audio piped to stdout and saved, then synced in Blender.

Radar WGS84 coordinates: 45.286757,14.202732 http://www.panoramio.com/photo/26952908