Note: Descriptions are shown in the official language in which they were submitted.
SYSTEM AND METHOD FOR CELLULAR NETWORK IDENTIFICATION
TECHNICAL FIELD
[0001] This invention relates generally to signal processing for wireless
communication
systems, and more particularly to a method and system for identifying signals
from different
cellular networks, such as GSM, CDMA2000, UMTS and LTE cellular networks.
BACKGROUND
[0002] Automatic signal identification (ASI) is an essential part of
intelligent radios used
in various military and commercial applications, such as electronic warfare,
spectrum
surveillance and software-defined and cognitive radios.
[0003] ASI algorithms can be categorized into two main classes: likelihood-
based and
feature-based. While the former algorithms provide the maximum average
probability of correct
identification, they are complex to implement and sensitive to model
mismatches. On the other
hand, feature-based algorithms are in general simpler to implement and robust
to model
mismatches. However, their performance is sub-optimal. In the prior art,
feature-based
algorithms have typically been used, and then only to identify standard
signals and most of them
concern the identification of a single standard signal (i.e., signal versus
noise, also known as
signal detection). For the identification of cellular standard signals versus
other signals, second-
order cyclostationarity-based features have been used to identify IEEE 802.11
standard signals,
Long Term Evolution downlink (LTE-DL) versus WiMAX signals, GSM versus LTE-DL
signals, and GSM versus CDMA and orthogonal frequency division multiplexing
(OFDM)
signals. A wavelet-based algorithm has been used to identify GSM versus CDMA
signals, with
GSM employing Gaussian minimum-shift-keying (GMSK) modulation and CDMA using
offset
quadrature phase-shift-keying. However, the prior art in general requires long
observation
intervals, which may not be available in certain applications. For example,
only a small portion
1
CA 2977226 2017-08-25
of the jamming interval is assigned for signal captures in reactive jamming in
order not to
degrade jamming performance, which leaves only short durations for signal
identification.
[0004] Accordingly, there is a need for improved cellular network
identifications
methods and/or systems.
SUMMARY OF THE INVENTION
[0005] In one embodiment of the invention, there is disclosed a method for
identifying cellular networks using a computer processor and a signal receiver
comprising:
[0006] (i) applying a signal magnitude test to determine whether magnitude
of
observed cellular signal samples follows a predetermined distribution
representative of
LTE-DL (long term evolution downlink) signals; upon a condition in which the
observed
samples follows the predetermined distribution, outputting an indication that
the signal is
from an LTE-DL network and terminating the method;
[0007] (ii) applying a test to determine the presence of a cyclic prefix;
upon a
condition in which the presence of a cyclic prefix is detected, outputting an
indication that
the signal is from an LTE-UL (long term evolution uplink) network, and
terminating the
method;
[0008] (iii) applying a test to determine whether the signal samples have
a constant
envelope; upon a condition in which the signal samples have a constant
envelope, outputting
an indication that the signal is from a GSM network, and terminating the
method;
[0009] (iv) applying a test to determine a signal bandwidth estimate to
determine
whether the signal samples are from CDMA2000 (code-division multiple access)
or UMTS
(universal mobile telecommunications system) networks; and outputting an
indication of
either the CDMA2000 or UMTS networks.
2
CA 2977226 2017-08-25
[0010] In one aspect of the first embodiment, the magnitude test is a
Kolmogorov-
Smirnov test, and wherein the predetermined distribution is a Rayleigh
distribution.
[0011] In another aspect of the first embodiment, the test to determine
the presence of
a cyclic prefix is a second-order one-conjugate correlation test.
[0012] In another aspect of the first embodiment, the test to determine
whether the
signal samples have a constant envelope is a one-sample Kolmogorov-Smirnov
test.
[0013] In another aspect of the first embodiment, a bandwidth of 1.25 MHz
is used to
indicate the CDMA2000 network and a bandwidth of 5 MHz is used to indicate the
UMTS
network.
[0014] In a second embodiment of the invention, there is provided a method
for
identifying cellular networks using a computer processor and a signal receiver
including:
[0015] (i) applying a test to determine whether observed cellular signal
samples have
a constant amplitude; upon a condition in which the observed samples have a
constant
amplitude, outputting an indication that the signal is from a GSM network;
[0016] (ii) applying a test to determine the presence of a cyclic prefix;
upon a
condition in which the presence of a cyclic prefix is detected, determining a
peak to average
power ratio of the signal; wherein a high peak to average power ratio is
indicative of an
LTE-DL signal and a low peak to average power ratio is indicative of an LTE-UL
signal;
and outputting a corresponding indication of either the signal is from either
an LTE-DL or
LTE-UL network;
[0017] (iii) applying a test to determine a signal bandwidth estimate to
determine
whether the signal samples are from CDMA2000 (code-division multiple access)
or W-
CDMA networks; and outputting an indication of either the CDMA2000 or W-CDMA
networks.
3
CA 2977226 2017-08-25
[0018] In a third embodiment of the invention, there is disclosed a system
for
identifying cellular networks including a signal receiver configured to
receive a cellular
signal and a a computer processor configured to receive data corresponding to
the cellular
signal and to carry out the method of either the first or second embodiments.
[0019]
BRIEF DESCRIPTION OF THE DRAWINGS
[0020] The invention is illustrated in the figures of the accompanying
drawings which
are meant to be exemplary and not limiting, in which like references are
intended to refer to like
or corresponding parts. Acronyms used in the description of the drawings are
expanded upon in
the description. Below is a listing of the drawings:
[0021] FIG. 1 shows a decision tree for the cellular network
identification method in one
embodiment of the invention.
[0022] Fig. 2 shows the absolute value of the measured second-order zero-
conjugate
function for the SC-FDMA signal as determined during the identification of an
LTE-UL signal.
[0023] Fig. 3 shows the mean square error with and without applying the
wavelet-based
denoising algorithm used in identifying a GSM signal.
[0024] Fig. 4 shows the CDF of the observed samples and the Rician CDF
with estimated
parameters.
[0025] Fig. 5 shows the effect of the number of captured samples, N, on
the percentage
of correct identification for LTE-DL and UMTS signals.
[0026] Fig. 6. shows the effect of the number of captured samples, N, on
the percentage
of correct identification for LTE-DL, UMTS and CDMA2000 signals.
4
CA 2977226 2017-08-25
[0027] Fig. 7 shows the effect of the number of captured samples, N, on
the percentage
of correct identification for LTE-DL, UMTS, CDMA2000, GSM and LTE-UL signals.
[0028] Fig. 8 shows the effect of the number of observed intervals, NB, on
the percentage
of correct identification for LTE-DL and UMTS signals under a fading channel.
[0029] Fig. 9 shows the percentage of correct identification for LTE-DL,
UMTS and
CDMA2000 signals under a fading channel.
[0030] Fig. 10 shows the percentage of correct identification for LTE-DL,
UMTS,
CDMA2000, GSM and LTE-UL signals s fading channel.
[0031] Fig. 11 shows an alternative embodiment of the method of Fig. 1.
DETAILED DESCRIPTION OF THE INVENTION
[0032] A practical example of the invention resides in its ability to
improve the
efficiency of a jammer. While a traditional blind signal jammer uses a noise-
like signal to
neutralize all types of threats, a more efficient waveform can be designed if
the target signal is
known. This requires the jamming system to have sensing capabilities for
identifying the type of
the target signal and adapt the jamming waveform accordingly. For example, in
advanced
electronic attack (EA), a jammer can synchronize to the threat network and
transmit a specialized
waveform for that practical network type to defeat the threats more
affectively. However, this
type of attack requires the knowledge of the network type, which may not be
known a priori in
ad hoc deployments, as different cellular network technologies can exist in
the same frequency
bands at different geographical locations. Our invention does not rely on
contiguous sample
captures for signal identification. In the state of the art of signal
identification, as outlined in
documents such as ITU-R SM. 1600-1, "Technical identification of digital
signals" , it is custom
to use correlation based techniques where received signal is correlated with
known preambles
embedded within the transmitted burst to detect/identify signals. However,
these preamble
signals are only transmitted during a small portion of longer bursts. If the
receiver is only
CA 2977226 2017-08-25
capturing small signal segments separated by longer intervals where there is
no signal
acquisition, it is very likely that the receiver will not capture preambles,
and state-of-art
techniques based on correlation will fail to identify the signal. Our
technique does not rely on a
priori preamble knowledge; therefore it is immune to interrupted data
acquisition problem.
[0033] In this invention, feature-based approaches are developed to
identify cellular
networks. In particular, the following identification features are employed.
[0034] The cumulative distribution function (CDF) of the magnitude of the
observed
samples provides a significant identification feature for the GSM and LTE-DL
networks. One-
sample and two-samples Kolmogorov-Smirnov (KS) tests can be used as decision
criteria in
different embodiments of the invention.
[0035] The signal bandwidth is used to differentiate between cdma2000 and
UMTS
signals, i.e., signals from wide-band code division multiple access (WCDMA)
networks in one
embodiment of the system.
[0036] Peak-to-average power ratio is used to differentiate between LTE-UL
and LTE-
DL signals. The existence of cyclic prefix (CP) is exploited to differentiate
between CDMA and
OFDM type modulations to identify LTE and CDMA2000/W-CDMA systems.
[0037] In the present invention, there is disclosed a general
identification method that can
be implemented on intelligent radio devices to identify LTE, UMTS, GSM, and
CDMA2000
networks. Novel identification features based on the statistics and the
structure in time and
frequency domain of the candidate signals are disclosed. The cumulative
distribution function
(CDF) of the magnitude of the observed samples is employed as an
identification feature for the
LTE-DL and GSM signals, and the decision criterion is designed based on the
Kolmogorov¨
Smirnov (K-S) test. Furthermore, the presence of the cyclic prefix (CP) and
the signal bandwidth
is exploited to identify the LTE uplink (LTE-UL), and UMTS versus CDMA2000
signals,
respectively. The method is extended to efficiently work under frequency-
selective fading
6
CA 2977226 2017-08-25
conditions by considering multiple non-contiguous observations and combining
the extracted
identification features from these observations.
[0038] The proposed identification method is intended for intelligent
radios with a single
receive antenna. The radio can be implemented in a reactive jammer device or a
receiver in a
heterogeneous network, where the cellular network type in a given channel
needs to be
identified. The general model used in this invention for the received signal
is y, = x, + Wit, n = 0,
1, . . . , N-1, where xn represents the transmitted signal which belongs to
either LTE-DL, GSM,
LTE-UL, UMTS, or CDMA2000 networks, N is the number of observed samples, and
wn is the
complex additive white Gaussian noise (AWGN) with zero-mean and variance o-,2.
[0039] Fig. 1 summarizes the four stages for the proposed cellular network
identification
method. In Stage 1, the one-sample K-S test is applied to check whether the
magnitude of the
observed samples follows a Rayleigh distribution, which is a unique feature of
the LTE-DL
signal versus the other signals under investigation. If the magnitude of the
observed samples
does not follow a Rayleigh distribution, the second-order one-conjugate
correlation test is
applied in Stage 2 to detect the presence of the CP. The LTE-UL network is
selected if a
significant peak is detected at delay corresponding to the useful time
duration of the single-
carrier frequency division multiple access (SC-FDMA) block, T, = 66.67 ps. In
the absence of
such a peak, the one-sample K-S test is applied in Stage 3 to check if the
received samples have
a constant envelope, which represents a characteristic feature for the GSM
signal. Finally, if no
such feature is exhibited, the signal bandwidth estimate is used in Stage 4 to
decide whether the
observed samples belong to CDMA2000 or UMTS networks. Note that the design of
the
proposed method and the choice of the identification feature at each stage is
done as shown in
Fig. 1, in order to avoid the use of common features. For example, at Stage 1,
we employ a
unique feature of the LTE-DL before applying the CP detection at Stage 2,
which is a common
feature of the LTE-DL and LTE-UL signals. A detailed description of each stage
now follows.
[0040] Stage 1: Identification of LTE-DL
7
CA 2977226 2017-08-25
[0041] OFDM modulation is employed for the LTE-DL signal, and thus, based
on the
central limit theorem, the LTE-DL signal has a complex zero-mean Gaussian
distribution.
Hence, the magnitude of the LTE-DL signal, zn = [ynl, follows a Rayleigh
distribution with the
CDF expressed as PLTE)(z) = 1 ¨ exp (¨z2/o-y2 ), where ay2 is the variance of
the received signal.
Note that cry2 y can be estimated from the observed samples as ê=
[0042] The identification of the LTE-DL signal relies on checking whether
the empirical
CDF of the magnitude of the observed samples, zn, n = 0, 1, . . . , N ¨ 1,
follows the Rayleigh
CDF. This can be done using the one-sample K-S test, which is a nonparametric
goodness of fit
test used to approve the null hypothesis that two data populations are drawn
from the same
distribution to a certain level of significance. Failing to approve the null
hypothesis indicates that
the two data populations follow different distributions. The hypothesized CDF
of the magnitude
of the LTE-DL signal can be calculated from the observed samples. As such, the
one-sample K-S
test can be used to identify the LTE-DL based on the test statistic D(I-TE) =
maxzn I (zn)) ¨
PLTE)(Zn))1, where f-'1 (zn)) is the empirical CDF calculated from the
observed samples. If D(LTE) <
y, where y is a predetermined threshold, the signal located in the target
frequency band is decided
to be LTE-DL.
[0043] It is worth mentioning that as the identification of the LTE-DL
relies on the CDF
of the magnitude of the observed samples, the carrier frequency offset does
not affect the
identification feature. Moreover, any delay in the observed samples does not
affect the
distribution of the magnitude of the signal. Accordingly, the identification
of the LTE-DL signal
is robust to carrier frequency and timing offsets.
[0044] Stage 2: Identification of LTE-UL
[0045] SC-FDMA is used for the UL transmission in LTE networks. For SC-
FDMA, a
CP of length v is added to each transmitted block by appending the last v
samples as a prefix.
This feature is used to identify the LTE-UL signals. To this end, we employ
the second-order
8
CA 2977226 2017-08-25
one-conjugate correlation R,(7) = E{yriy,*+õ-} to detect the CP in the
observed samples, where
E{.} and * denote the statistical expectation and complex conjugate,
respectively. The
correlation (2,(T) is estimated as
N-1
1
fic(r) = YnYn* +I- = (1)
n=o
[0046] In the presence of the CP,1 f?',(1)1 exhibits a peak at a delay
corresponding to the
useful symbol duration, 71, i.e., Tp = pNc, where p is the oversampling factor
and ATc is the number
of subcarriers. Therefore, the presence of such a peak indicates that the
cellular network within
the target band is LTE-UL. Fig. 2 shows the absolute value of the measured
second-order zero-
conjugate correlation function 1 f?c(r) 1 for the SC-FDMA signal with N =
4000, N, = 128, p = 4,
and long CP, i.e., v = Nc/4, over AWGN channel at SNR = 6 dB.
[0047] Note that Rc(T >> Tp) is theoretically zero when N co. However,
with a limited
number of observed samples N, its value is nonzero and represents the
estimation error.
[0048] Such an error has an asymptotic zero-mean complex Gaussian
distribution with
variance c)-. Accordingly, the normalized correlation function R(T) = (2/o-
,2)1/2 f?(7)r>> Tp,
follows a zero-mean complex Gaussian distribution with variance equal to two;
hence, 1 fi(c)12, T
>> Tp, has a chi-square distribution with two degrees of freedom. Accordingly,
we define the test
statistic, F as r = 2/0 c(T)12, with (5-',2 =1/11 Er Tp I fic (T) 12
, where SI is the cardinality of the
set of considered delays r)> Tp.
[0049] Furthermore, if F? i, the LTE-UL network is decided to be present
in the target
frequency band. The threshold i corresponds to the desired probability of
false alarm, Pi-, = Pr(F
> q1LTE network is not present). Based on the CDF expression of the chi-square
distribution
with two degrees of freedom, i = ¨2 ln Pf a.
9
CA 2977226 2017-08-25
[0050] Note that as the identification of LTE-UL depends on the magnitude
IR,,(1-)1 at a
certain delay, Tp, its identification is robust to carrier frequency and
timing offsets.
[0051] Stage 3: Identification of GSM
[0052] The GMSK modulation is inherited in the GSM signal; as such, the
transmitted
signal, xn, has a constant envelope, A Vn, where A is a positive constant.
Given that wn
Nc(0, oZ), the magnitude of the received GSM signal, zn = + wn1 follows a
Rician distribution
whose CDF is expressed as PGsm)(z) = 1 ¨ Q1((f1Al/o-w), (V2z/o-w)),where
Q1(.,.) is the
Marcum-Q function.
[0053] Similar to LTE-DL, the identification of the GSM signal relies on
checking
whether the empirical CDF of the magnitude of the observed samples, zn, n = 0,
1, . , N ¨ 1,
follows the Rician CDF. Given the GSM signal amplitude, A, and the noise
variance, ad,, the
one-sample K-S test can be applied to identify the GSM signal; in practice,
these two parameters
need to be estimated. The GSM signal amplitude can be estimated as A =
(11N)EõN=c1 zn.
Furthermore, the variance ê can be estimated as 6,2 = (2/N) EnNIO Al2.
[0054] For a more accurate estimate of A, a wavelet-based denoising
algorithm is applied
to the observed samples, y1, before calculating A. This can be done by
applying a wavelet
transform to the observed samples, zeroing all the resulting wavelet
coefficients below a
predetermined threshold, and finally applying an inverse wavelet transform
process to obtain the
denoised signal. Fig. 3 shows the mean square error (MSE) of A, i.e., MSE =
EllA ¨ A121, with
and without using the denoising algorithm. Additionally, Fig. 4 shows the CDF
for the observed
samples and the Rician CDF with the estimated parameters, A and 6,2, with N =
8000 and at
SNR = 10 dB. Now, the test statistic D(GSM) = maxzn1 Pi(zn)) ¨ (p osn4)(
z ))iis used to identify
the GSM network. If D(GSM)<y, where y is a predetermined threshold, the signal
is decided to be
GSM.
CA 2977226 2017-08-25
[0055] Similar to the LTE-DL identification, the CDF of the magnitude of
the observed
samples is employed to identify the GSM signal. Therefore, the identification
of the GSM signal
is also robust to carrier frequency and timing offsets.
[0056] Stage 4: Identification of UMTS and CDMA2000
[0057] The UMTS and CDMA2000 signals are both generated using direct
sequence
spread spectrum, and the signal bandwidth represents the main difference
between the two
networks, i.e., a bandwidth of 1.25 MHz is used for CDMA2000, whereas 5-MHz
bandwidth is
assigned to UMTS. This means that by estimating the signal bandwidth, we can
differentiate
between the UMTS and CDMA2000 signals. We employ the Welch's method to
estimate the
power spectral density of the observed samples. This is done by splitting the
observed samples
into overlapping segments and applying windowing and fast Fourier transform on
each segment.
The power spectral density is calculated on each segment, with the average
being the estimated
power spectral density of the observed samples. The estimated signal
bandwidth, BW, is
calculated from the estimated power spectral density as the difference in
frequency between the
points where the integrated power crosses 0.5% and 99.5% of the total power in
the spectrum. If
IBW ¨1.25 MHz' > IB W ¨5 MHzI, then UMTS is selected; otherwise, CDMA2000 is
identified
as the observed cellular network. Note that the carrier frequency and timing
offsets do not affect
the estimation of the signal bandwidth. As such, the identification of the
UMTS and CDMA2000
signals is robust to such impairments.
[0058] Identification in Fading Conditions With Noncontiguous Intervals
[0059] In practice, it is possible that NB noncontiguous observation
intervals are available
to be used for identification, which can be beneficial under fading
conditions. Each observed
interval contains N samples, which are considered to experience independent
fading. In such a
case, the identification features are combined as follows.
11
CA 2977226 2017-08-25
[0060] For the GSM and LTE-DL signals, the K-S test is applied on N
samples at each
observed interval. The majority rule is applied to decide on the selected
cellular network. For
example, if D(LTE)<y for more than NB/2 intervals, the LTE-DL is decided to be
the signal located
within the target frequency band. The same process is applied to identify the
GSM signal.
[0061] For the identification of the LTE-UL signal, the second-order zero-
conjugate
correlation is estimated over each interval and the test statistics on each
observation interval, Fb,
b = 0, 1, 2, . . . , NB ¨ 1 are combined as F =ENb B0-1 rb. If r > 77, then
the present signal is LTE-
UL. Note that the null hypothesis in this scenario has a chi-square
distribution with the degree of
freedom equal to 2NB, and the threshold 77 is set based on that distribution
and the desired Pfa .
[0062] For the identification of the UMTS and CDMA2000 signals, the
estimated power
spectral density is averaged over the observed intervals; then, the signal
bandwidth is estimated
based on this average.
[0063] EXPERIMENTAL RESULTS
[0064] Measurement Setup
[0065] The proposed method is evaluated using frequency division duplex
standard
cellular signals generated by a Rohde&Schwarz (R&S) SMU200A vector signal
generator and
acquired with an R&S FSQ8 signal analyzer. The waveform generator and signal
analyzer are
not synchronized; therefore, the captured samples are affected by unknown
timing and frequency
offsets. Acquired signals are captured from the IF output of the signal
analyzer and then
transferred to a personal computer over the Ethernet, where the identification
process is done
using MATLAB. The AWGN and ITU-R vehicular A channels are considered to
evaluate the
performance of the identification method. The maximum delay spread for the ITU-
R vehicular A
channel is set to 2.51 us. The maximum Doppler frequency is set to 145.69 Hz.
The average
energy of the received samples is normalized to unity, and the SNR is defined
as SNR = 10
log io(1/o-,2),where oZ represents the variance of the noise samples. Unless
otherwise mentioned,
12
CA 2977226 2017-08-25
the number of observed samples, N, is 4000, the level of significance used
with the K-S test isle'
= 0.999, and the probability of false alarm, PI", is 10-3. Three different
sets of candidate signals
are considered: 01 = ILTE-DL,UMTSI, 02 = ILTE-DL,UMTS,CDMA20001, and 03 = {LTE-
DL,UMTS,CDMA2000,GSM, LTE-UL}. Unless otherwise mentioned, the LTE-DL and LTE-
UL signals with a bandwidth of 1.4 MHz, i.e., 128 subcarriers, are used. The
percentage of the
correct identification is used as a figure of merit, and calculated based on
1000 Monte Carlo
trials.
[0066] Method Validation Under AWGN Channel
[0067] Fig. 5 shows the effect of the number of captured samples, N, on
the identification
for 01 over AWGN channel. Note that for 01, the one-sample K-S test (Stage 1
in Fig. 1) is
applied to identify the LTE-DL signal versus the UMTS signal. Clearly, the
proposed
identification method provides perfect identification for the LTE-DL signal
regardless of the
SNR level. This can be explained, as the Gaussian noise does not affect the
distribution of the
LTE-DL signal, and the identification feature still exists regardless of the
SNR value. On the
other hand, identification of the UMTS signal enhances by increasing N and
perfect
identification can be achieved with N = 6000 at SNR = 0 dB.
[0068] For 02, the identification process relies on the one-sample K¨S
test (Stage 1) and
bandwidth estimation (Stage 4). Fig. 6 shows the identification performance of
02 with different
number of captured samples, i.e., N = 2000, 4000, 6000, over AWGN channel. As
previously
noticed, perfect identification of the LTE-DL signal is obtained regardless of
the SNR. For the
UMTS and CDMA2000 signals, the proposed identification method works perfectly
at SNR = 4
dB with N = 2000, while perfect identification is attained with N = 4000 at
SNR = 2.With N =
6000, the UMTS and CDMA2000 signals are identified perfectly at SNR = 0 dB and
SNR = 2
dB, respectively.
13
CA 2977226 2017-08-25
[0069] Fig. 7 shows the identification performance of 03, over AWGN
channel with N =
4000 and N = 6000. As expected, the identification performance of the LTE-DL
signals does not
depend on SNR. On the other hand, the LTE-UL signal has the lowest performance
as the
discriminating peak, I fl,(Tp)I is highly affected by the noise level and does
not exist at low SNRs.
Overall, a perfect identification performance is achieved for all candidate
cellular networks with
N = 4000 and N = 6000 at SNR > 5 dB and SNR > 4 dB, respectively.
[0070] Performance Comparison
[0071] Table I compares the identification performance of the proposed
method with the
techniques used the prior art references identified below with N = 4000 over
AWGN channel at
SNR = 2 dB. As can be noticed, the proposed identification method outperforms
the three
techniques. Such identification features require a large number of samples to
achieve good
identification performance. This performance highlights the advantage of the
method of the
invention in providing a good identification performance with a short
observation interval.
TABLE I
PERCENTAGE OF CORRECT IDENTIFICATION OF THE SIGNAL A FOR THE PROPOSED METHOD
AND
THE TECHNIQUES IN [1]¨[3] WITH N = 4000 OVER AWGN CHANNEL AT SNR =2 dB
= GSM 2= LTE-DL 2 =CDMA2000
Proposed 100 100 100
algorithm
Algorithm in 2.12 8.12
111
Algorithm in 4.89 3.12 5.13
[2]
Algorithm in 52.3 100
[31
[1] E. Karami, 0. A. Dobre, and N. Adnani, "Identification of GSM and LTE
signals using their second-order
cyclostationarity," in Proc. IEEE IMTC, May 2015, pp. 1108-1112.
[2] M. oiler and F. Jondral, "Cyclostationarity based air interface
recognition for software radio systems," in Proc.
IEEE RWC, Sep. 2004, pp. 263-266.
[3] B. Liu and K. C. Ho, "Identification of CDMA signal and GSM signal using
the wavelet transform," in Proc. IEEE
MSCS, vol. 2. Aug. 1999, pp. 678-681.
14
CA 2977226 2017-08-25
[0072] Method Validation Under Fading Channels
[0073] In this section, we present the effect of the ITU-R vehicular A
fading channel on
the identification performance of the proposed method. In this scenario, the
decision on the
present cellular network is made based on NB noncontiguous observation
intervals, which exhibit
independent fading conditions, and N samples are captured during each
interval.
[0074] Fig. 8 shows the effect of the number of observed intervals on the
percentage of
correct identification for 01 under ITU-R vehicular A fading channel. Each
observed interval
contains N = 2000 samples. As previously mentioned, perfect identification of
the LTE-DL
signal is obtained regardless of the number of observed intervals. On the
other hand, the
identification performance of the UMTS signal enhances as NB increases. This
is because the
fading channel can have a destructive effect on the identification feature,
which is reduced using
different intervals with independent channel parameters.
[0075] Fig. 9 shows the percentage of correct identification for 02 under
ITU-R vehicular
A fading channel with NB = 5 and N = 2000. The proposed method provides
perfect
identification performance at SNR = 10 dB. It is worth mentioning that, as
previously presented
for other cases, better performance can be achieved by increasing the number
of observation
intervals. The percentage of correct identification for 03 under ITU-R
vehicular A fading channel
with NB = 5 and N = 4000 is shown in Fig. 10. One can notice that better
performance is obtained
for the UMTS and CDMA2000 signals when compared with the GSM and LTE-UL
signals.
[0076] In conclusion, the proposed identification method relies on the
signal CDF and
the K-S test to identify the GSM and LTE-DL signals. Furthermore, the presence
of the CP in
LTE-UL signal is exploited as an identification feature. Finally, the
bandwidth measurement is
employed to identify the UMTS versus CDMA2000 signals. The identification
method is
evaluated using standard signals generated and acquired by an R&S vector
signal generator and
spectrum analyzer, respectively. Experimental results verify the validity of
the method with short
CA 2977226 2017-08-25
observation intervals under AWGN and ITU-R vehicular A channels. Moreover, the
method of
the invention is robust to timing and frequency offsets.
[0077] Various alternatives and variations to the invention are
contemplated, particularly
in respect of the decision tree, order of and determination made to
distinguish between types of
signals. One such variation is shown in Fig. 11.
[0078] Figure 11 shows a decision tree of an alternative embodiment of the
invention. In
this implementation, the first stage is to check if the signal has constant
amplitude. This check
can be performed using features described above. Because only GSM signals have
constant
amplitude among all cellular network signals, the first stage allows
discrimination between GSM
vs other cellular signals.
[0079] The second stage is to further discriminate between other cellular
signals. LTE is
the only cellular network standard that uses OFDM modulation, which relies on
cyclic prefix to
ease channel estimation. The test for the existence of cyclic prefix can
discriminate between LTE
vs other cellular signals. This check can be performed using, for example, a
second order
correlation algorithm used in the first embodiment of this invention.
[0080] In a successive stage of the decision tree, uplink vs downlink
identification of the
LTE network can be performed. The LTE downlink uses OFDMA scheme to ease
resource
allocation which generates large peak-to-average power ratio (PAPR) signals;
whereas low
PAPR SC-FDMA technique is used in the uplink to use hand-set battery more
efficiently. In this
embodiment of the system, PAPR of the received signal can be used to
discriminate between
LTE-UL and LTE-DL.
[0081] At the last stage of the decision tree, CDMA2000 vs W-CDMA
discrimination is
performed based on the bandwidth distinction of these two systems.
16
CA 2977226 2017-08-25
