![](bg3.png)
method is similar to Bartlett periodogram, the
difference lies in using D samples overlap of
segments and multiplying every segment by window
function w(n). Typical overlap values are 25%, 50%
or 75% of segment length L.
4.1.2 Multi Window Spectral Estimation
This method (sometimes called Multi tapper
method) uses the set of orthogonal sequences as the
windows applied to the periodograms. Final
spectrum estimate is given as the average of all
particular periodograms. The discrete prolate
spheroids - Slepian sequences are often used as the
windows.. The corresponding power spectrum
estimate is given as (Thomson, 1982):
()
∑∑
−
=
−
=
−
=
1
0
2
1
0
2
MW
)()(
11
m
i
N
n
fnj
ki
i
enxnw
m
fP
π
λ
(5)
Where m represents number of used windowing
sequences, w
i
and λ
i
are the i-th sequence and its
eigenvalue respectively.
4.1.3 Signal Key Features
This method initially proposed for the modulation
type classification has been proposed in (Ulovec,
2008). In that paper, several features (most of them
statistical moments) have been defined. During our
experiments, we have used mainly the feature
denoted A
S.
More informations about the method can
be found in the original paper. Defined signal key
features can be used also for modulation type
recognition. In TV bands, the classification into 3
classes – DVB-T, analog TV or noise is possible in
the assumption of sufficiently high SNR.
4.1.4 Cyclic Prefix Correlation
The correlation algorithm generally defined as:
)m()()( +=
∑
∞
−∞=
∗
ngnfnR
m
fg
(6)
can be used for signal presence detection. There are
several possible alternatives of signals f and g
assignment (autocorrelation, correlation of received
signal with known preamble etc.). We have used the
special property of the OFDM signal used for DVB-
T (ETSI, 2009) broadcasting – cyclic prefix. Sliding
correlation of two signals with duration of cyclic
prefix length and time separation corresponding to
the length of OFDM symbol useful part has been
used. Because both signals actually present in the
windows are almost identical, the peaks in the
correlation function will occur. The correlation peak
and average values have been measured and
compared with selected threshold.
4.2 Experimental Implementations
4.2.1 Sensing Device in Universal Software
Radio Peripheral
In this first described implementation, the radio
frequency signal has been received by the Universal
Software Radio Peripheral (USRP) device equipped
with TV tuner TVRX, both commercially available
from Ettus Reseach company. The received complex
baseband signal has been acquired to Simulink
environment, where the above mentioned sensing
methods have been tested. The issue of this
implementation for the use in European countries is
in different bandwidth of TV tuner developed for US
use.
4.2.2 PC-based Sensing Device by Sampling
DVB Tuner Output
This implementation has been based on the sampling
of the intermediate frequency output of
commercially available Humax F3-FOX T DVB-T
receiver. The received signal has been digitized by
Gage CompuScope 12400 card. The sampling
frequency of 100MHz with 12 bit resolution has
been used. The data have been subsequently
converted into baseband with the use of Hilbert
transformer and downsampled by factor of 10 in
order to relax the processing complexity.
4.2.3 FPGA Sensing Device by Sampling
DVB Tuner Output
The energy detector based on the periodogram was
synthesized for FPGA Virtex IV (device xc4vsx35)
device with use of the Xilinx System Generator
environment. For the real time implementation the
Memec Virtex IV MB Development Kit with
analogue module P240 was chosen (Memec design,
2005). Hardware details are described in the
following paragraph. The analogue module provides
dual channel analogue inputs and outputs. The A/D
converters are 14 – bits up to 125 MSPS. Similarly
to the previous case, the intermediate frequency
signal from DVB-T receiver has been sampled by
the A/D converters in analog module as is shown in
Figure 1.
EXPERIMENTAL EVALUATION OF RADIO FREQUENCY SPECTRUM SENSING DETECTORS IN TV BANDS
57