Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
!"#$%! & ' (
Abstract: - A method to detect and characterize a frequency hopping (FH) signal is proposed. The context is
Spectrum Surveillance. Hence, no a priori information is available: the parameters of the modulation must be
automatically determined. Spread spectrum (SS) modulations are well-known for their low probability of
interception (LPI). Nevertheless the aim of this paper is to show that, using a time–frequency representation and
ad-hoc eigenanalysis techniques, it is possible to characterize such signal, even in realistic conditions (low SNR).
Once characterized, the signal can be demodulated. Many tools are used for this characterization, in particular a
precise definition of characteristic parameters and mathematical methods, such as data compression, 3D rotations
and principal components analysis (PCA).
35
3.2 Sequence duration
A FH modulation is also characterized by the
30
spreading sequence duration Ds. This duration always
25 depends of Tf, but in our application it depends also
Frequency
[MHz] of the PN sequence length l.
20
Thus, we can write :
15 l *T f
Ds (1)
10
n
with n the power of 2 in the definition of the number
5
M of different symbols used in the spreading
0
5 10 15 20 25
modulation: M 2 n .
Time [µs]
The number of frequency stages is then defined by :
Ds
Ns (2)
Figure 1: Spectrogram (SNR=10dB) Tf
frequency extension
- The central frequency or carrier frequency f0 of the
30 signal.
frequency gap
- The frequency gap Δf between two adjacent
25
Frequency
[MHz]
frequencies.
20 frequency stability period - The frequency extension ext, which corresponds to
frequency bandwidth around a transmission
frequency.
15
10
0
2 4 6 8 10 12
and even to define the modulation type used for data
Time [µs]
symbols (ex : QPSK,…), and finally demodulate the
signal.
Figure 2: Parameters definition
!"#$%! & ' (
0
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
10
time.
50
60
5 Image compression
0
200 400 600 800 1000 1200 1400 1600 1800 2000
this new base. The matrix entry at row n and column 500
i is : 50
§ 2i 1 ·
I n i cos¨¨ nS ¸ (3)
60
2 N e 2 ¸¹
0
©
200 400 600 800 1000 1200 1400 1600 1800 2000
with 1 d i d N e where Ne is the number of columns of Figure 4: Original and reconstructed images
the image, and 1 d n d N where N is its number of
rows. This compression method consists in suppressing the
!"#$%! & ' (
highest image frequencies. When we look at C (fig. The method we developed consists in calculating the
5), we can see that the main energy is at the left of autocorrelation matrix of the rows of matrix C. We
the image, where the low frequencies are located. can observe the eigenvalues of the autocorrelation
This explains the low loss of information. matrix Rl C T C on figure 7.
4
x 10 9
1 x 10
10
0.8
10
9
0.6
8
20 0.4
7
0.2
6
30
0
5
-0.2
40 4
-0.4
3
50
-0.6
2
-0.8
60 1
-1
20 40 60 80 100 120 0
0 2 4 6 8 10 12 14 16 18 20
35 -0.04 -0.02
0 1000 2000 3000 4000 0 1000 2000 3000 4000
30
25
Figure 8: Rows analysis eigenvectors
Frequency
[MHz]
20
Each vector represents the temporal sequence of one
15
of the transmission frequencies or is a linear
combination of two or more sequences. Hence, in
10
order to remove linear combinations, we must rotate
5
the set of vectors until a criteria is minimized. Since
we know that two transmission frequencies cannot be
0
2 4 6
Time [µs]
8 10 12 present at the same time, the criteria is built in order
to reach a minimum when the following condition is
Figure 6: Spectrogram with SNR=-5dB true: “for each i, if the ith component of a vector is
non-zero, then the ith component of the other vectors
!"#$%! & ' (
¨ sin M cos M 0 ¸¹
© 1
where 0 d M d 2S and 0 d T d S . 0
0 2 4 6 8 10 12 14 16 18 20
The results are shown on figure 9. Figure 10: columns analysis eigenvalues
(20 first ones)
1st rows analysis eigenvector 2nd rows analysis eigenvector
0.06 0.06
Four large eigenvalues are clearly seen. They
0.04 0.04 correspond to the transmission frequencies. The
0.02 0.02 corresponding eigenvectors are displayed on figure
11.
0 0
-0.02 -0.02 1st columns analysis eigenvector 2nd columns analysis eigenvector
0 1000 2000 3000 4000 0 1000 2000 3000 4000 0.4 0.4
3rd rows analysis eigenvector 4th rows analysis eigenvector
0.06 0.08 0.3 0.2
0.06 0.2 0
0.04
0.04
0.02 0.1 -0.2
0.02
0 0 -0.4
0 0 20 40 60 80 0 20 40 60 80
-0.02 -0.02 3rd columns analysis eigenvector 4th columns analysis eigenvector
0 1000 2000 3000 4000 0 1000 2000 3000 4000 0.6 0.6
0.4
0.4
Figure 9: rows analysis eigenvectors after 0.2
3D-rotation 0
0.2
0
-0.2
After the rotation, each vector corresponds clearly to -0.4 -0.2
the temporal sequence of one of the transmission 0 20 40 60 80 0 20 40 60 80
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0 5 10 15 20 25 30 35 40 45
frequencies in MHz
7 Conclusion
The spread spectrum systems, and consequently
Frequency Hopping systems, are very difficult to
intercept. It is the main reason for which they are
used for many years for military applications. In this
paper we show that, by combining various analysis
techniques, it is really possible to characterize such
signals.
This new approach starts from a time-frequency
representation, and requires a precise definition of the
parameters, as well as the use of mathematical tools.
The method is able to demodulate signals in realistic
conditions, and even in bad conditions (SNR=-5dB).
Some techniques used could probably be adapted to
other digital modulations identification problems, and
this is part of our further work.
References:
[1] Savo Glisic, Branka Vucetic, Spread Spectrum
CDMA Systems for Wireless Communications,
Mobile Communications Series, Artech, 1997.
[2] R.Steele, Mobile Radio communications,
Pentech, 1992.
[3] F. Auger, P. Flandrin, P. Goncalvès, O. Lemoine,
Time-frequency Toolbox - For use with Matlab,
CNRS (France) / Rice university (U.S.A), 1996.
[4] John G. Proakis, Digital communications, Third
edition, McGraw-Hill, 1995.
[5] Marvin K. Simon, Jim K. Omura, Robert A.
Scholtz, Barry K. Levitt, Spread Spectrum
Communications, vol. III, Computer Science
Press, 1985.