Multi-Antenna LTE detection for Dynamic Spectrum Access: A Proof

Multi-Antenna LTE detection for Dynamic
Spectrum Access: A Proof of Concept
Nicola MICHAILOW1 , David DEPIERRE2 and Gerhard FETTWEIS3
1,3
Vodafone Chair Mobile Communications Systems
Technische Universität Dresden
01069 Dresden, Germany
2
THALES Communications and Security
Boulevard de Valmy 160
92700 Colombes, France
Email: 1 [email protected], 2 [email protected], 3 [email protected]
Abstract—Detection of occupied frequency bands is the foundation for applications of dynamic spectrum access (DSA). In
order to convince network operators that DSA is feasible in
cellular frequencies, it has to be shown that a reliable detection
of their primary signals is possible. In this paper, we present the
results of experimental validation of an algorithm and hardware,
which can detect the presence of a Long Term Evolution (LTE)
signal. In contrast to the classical mono antenna approach, an
array of antennas is used, which allows to enhance the detection
capabilities, particularly when besides the useful signal there is
also interference.
Index Terms—cognitive radio testbeds, spectrum sensing, feature detection, LTE
I. Motivation
Cognitive radio (CR) is a research area that has grown
significant popularity since it was introduced by Mitola [2].
While CR is not only about wireless communications but also
includes aspects from other areas like artificial intelligence and
machine learning, it is often used to address what is actually
dynamic spectrum access (DSA). DSA can be classified into
several categories of spectrum use: An exclusive use model,
an open sharing model and a hierarchical access model [3].
Reliable spectrum sensing is the foundation for coexistence of
wireless communications systems in the latter two of those
categories. There are three main methods for this process
[4]. In matched filter detection, coeherent demodulation is
performed. Thus the structure of the signal to-be-sensed needs
to be known. This method maximizes the signal-to-noise
ratio (SNR). Energy detection is a simplified, non-coherent
method that is typically based on the fast Fourier transform
algorithm. It is however susceptible to varying noise levels
and interference. Finally, the cyclostationary feature detection
is a sophisticated approach that relies on periodicity in the
transmit signal, e.g. cyclic prefixes or frame structures.
Further, the sensing process can be enhanced by using an
antenna array [1]. Multi-antenna reception and antenna processing can be of great benefit in sensing. It allows to improve
the detection performance by several means: The detection is
more robust to fast fading thanks to spatial diversity. If the
level of the received signal is very low at one given antenna
due to a deep fade, the signal level on the neighbor antenna
might not be subject to such fading. With a distance of one
wavelength apart, the phases of a propagation channel can
be considered independent, provided that the angle spread is
sufficiently high at the receiver. Additionally, signals from
different spatial directions can be separated. This allows to
detect a given source even with very low signal-to-interference
ratio (SIR). When the antenna array receives several sources
emitted from different locations, the antenna processing can recombine the signals in order to detect one given source, while
the other are considered as interference which is spatially
rejected. The presented techniques must not be mistaken with
distributed sensing. Here, the sensors are co-located roughly
one wavelength apart. This kind of setup is typically referred
to as single input multiple output (SIMO), i.e. a system with
one antenna at the emitter and several antennas at the receiver.
The rest of this paper is organized as follows. In section 2
multi-antenna sensing in cellular networks is motivated with
two use cases. In section 3, the mathematical background
of the detection algorithm is presented. Section 4 contains a
description of the hardware platform, the experiment setup
as well as the results. Section 5 presents how this proof of
concept is related to cognitive radio experimentation in the
CREW project. In section 6, conclusions are drawn.
II. Use Cases for LTE Detection
In this work, a hardware platform for the detection of Long
Term Evolution (LTE) downlink signals is presented, which
operates based on matched filter multi-antenna detection using
the primary and secondary synchronization sequences (PSS
Fig. 1. Metrology oriented use case: Interference analysis based on smart
antenna processing
and SSS) as a reference. This approach is particularly relevant
for two use cases:
The platform can be employed as a metrology tool for
network planning, as it allows downlink interference analysis
of LTE cellular networks. The scenario is depicted in Fig.
1. The process consists of creating a list of the received
signals on an LTE channel as well as obtaining their physical
characteristics and identifying the base stations which are
transmitting these signals. Interference analysis is based on
the smart antennas approach employing an array of antennas.
When this antenna array receives several sources emitted by
different base stations, the signals are recombined in order to
detect and demodulate each received source, while the other
received signals are considered as interference that can be
rejected when it comes from different directions. Since the
matched filter approach is used for detection, the signal can
be demodulated and system parameters like the physical layer
cell identity, cyclic prefix length, duplex mode, base station
signal level, Ec /I0 , as well as time and frequency channel
impulse response can be determined. This functionality can
be used by a network operator, either to optimize the network
as co-channel interference due to frequency reuse can never
be completely avoided in cellular systems, or to complain to
the regulator about interference systems occupying adjacent
frequency bands.
LTE detection with spatial interference rejection capabilities
can also be useful for sensing in CR environments. Consider
the scenario depicted in Fig. 2: A mobile opportunistic receiver
(Rx) uses a white space in a certain frequency band that is
available at a particular geographic location (a). The system
has to check periodically for the presence of an incumbent
user that might be utilizing that band. When the mobile
opportunistic Rx moves to a location where the white space is
no longer available, it receives the signal from the incumbent
LTE transmitter (Tx). While the opportunistic system is not
aware of the presence of the incumbent Tx, it is still communicating and for the opportunistic Rx, the signal coming
from the incumbent Tx is considered as interference (b). When
the opportunistic Rx senses the spectrum, the signal coming
from the incumbent LTE Tx is now the signal of interest
and the signal from the opportunistic Tx the interference,
as it is jamming the sensing process (c). Typically, during
this process silent periods are allocated in the opportunistic
Fig. 2.
Cognitive radio oriented use case: Primary user detection with
interference rejection
system waveform, in order to perform sensing without being
jammed by its own system. But sensing can also be performed
simultaneously with the data transfer, if the opportunistic Rx
has interference rejection capabilities, i.e. by using an antennaarray and antenna processing algorithms.
III. Theoretical Background
Let
 1 
 x [n] 


x[n] =  ... 
 M 
x [n]
(1)
be a vector that contains snapshots of the received signal
of an array with M antennas, where xm [n] denotes the time
signal received at time n on antenna m. Presence or absence
of the synchronization signal at time n can be formulated as
a composite hypothesis-testing problem with the two options
Hypothesis H0 ”Absence of the synchronization signal” and
Hypothesis H1 ”Presence of the synchronization signal”. Under H0 , the discrete time signal x can be written as
x[n + k] = n[n + k],
while under H1 , it is denoted by
X
x[n + k] = hd[k] +
h p d[k − p] + n[n + k].
(2)
(3)
p,0
Here, n[n] represent the contribution of background noise,
P
h + p,0 h p δ[n − p] is the unknown transfer function of the
discrete time propagation channels between the transmitter
and the M receive antennas and d[k], k = 0, . . . , N − 1 is the
synchronization sequence of length N.
A. Optimal Spatial Detector
The the joint probability distribution of the sequence
{x[n + k], k = 0, . . . , N − 1} depends on several unknown parameters. It is thus not possible to derive optimum detection
procedures in the most general case. In order to motivate
the use of sub-optimum algorithms, we first address the case
552 samples
1
CP
40 samples
1
CP
2
C
P
2
36 samples
3
C
P
3
C
P
4
C
P
5
C
P
6
512 samples
2
CP
128 samples
1
LTE slot : 0.5 ms
3840 samples (for 5MHz bandwidth)
548 samples
C
P
CP
4
5
CP
6
CP
512 samples
4
5
6
slot
Extended CP
PSS
7
8
where m = 0, . . . , Nfr − 1. The ML test then consists of
comparing the quantity
Normal CP
640 samples
3
CP
7
9
10
11
12
13
14
15
16
17
18
19
c̄(n) =
SSS
20
LTE frame structure and synchronization sequences in FDD mode
c̄(n) =
where the noise n[n] is temporally white and the channel
transfer function is reduced to h, which implies a single path
propagation channel between each active base station and the
receiver. In that case, the received signal under H1 can be
written as
x[n + k] = hd[k] + n[n + k].
(4)
With these assumptions, it is possible to derive the maximum
likelihood ratio test as
!N
det (R0 )
...
c(n) =
det (R1 )
 PN−1
 (5)
 − k=0 (x[n + k] − hd[k]) R−1

1 (x[n + k] − hd[k]) 

 ,
exp 
PN−1
−1 H
− k=0 x[n + k]R0 x [n + k]
where R0 and R1 are the unknown covariance matrices of the
noise under H0 and H1 respectively.
Together with the vector h they have to be estimated under
each hypothesis in the maximum likelihood (ML) sense. Under
H0 , R0 can be estimated as R̂0 which is given by
R̂0 = R̂ xx [n],
(6)
while under H1 the estimates for h and R1 are
1
ĥ =
r̂ xd [n]
(7)
||d||2
and
1
R̂1 = R̂ xx [n] −
r̂ xd [n]r̂Hxd [n].
(8)
||d||2
PN−1
PN−1
Therein, ||d||2 = k=0
|d[k]|2 , r̂ xd [n] = k=0
x[n + k]d∗ [k] and
PN−1
H
R̂ xx [n] = k=0 x[n + k]x [n + k].
By using (6), (7) and (8) in (5), the ML ratio is obtained as
r̂ xd [n]H R̂−1
xx [n]r̂ xd [n]
.
||d||2
B. Use of Periodicity
c(n) =
(9)
To improve the performance of the test, the periodicity
of the synchronization sequences can be exploited. For that
purpose, the hypothesis-testing problem has to be modified to
take into account for the number of observed frames Nfr . With
a frame duration T fr , H0 can be rewritten as
x[n + mT fr ] = n[n + mT fr ]
(10)
and H1 as
x[n + mT fr ] = hd[k] + n[n + mT fr ],
(12)
to a threshold. If all c(n + mT fr ) are small compared to 1, the
expression can be reduced to
LTE frame: 10 ms
20 slots
76800 samples (5MHz mode)
Fig. 3.
Nfr −1
1 X
−ln (1 − c(n + mT fr ))
Nfr m=0
(11)
Nfr −1
1 X
c(n + mT fr ).
Nfr m=0
(13)
Thus, c(n) can be seen as an instantaneous criterion, while
c̄(n) constitutes an averaged criterion.
C. Application to LTE
The downlink parameters of LTE are summarized in Table
I for the 5 MHz and 20 MHz mode. The symbols are grouped
Parameter
Mode 1
Mode 2
channel bandwdith
5 MHz
20 MHz
512
2048
15 kHz
15 kHz
FFT size
subcarrier bandwidth
occupied subcarriers
300
1200
occupied bandwidth
4.515 MHz
18.015 MHz
sampling bandwidth
7.68 MHz
30.72 MHz
TABLE I
Relevant LTE downlink parameters
in slots with a duration of 0.5 ms. One frame consists of 20
slots, which corresponds to 10 ms. The number of symbols
in a slot depends on the length of the cyclic prefix (CP): In
normal CP mode, one slot contains 7 symbols and the length
of the first symbol’s CP is longer in order to keep the slot
duration constant. In extended CP one slot contains 6 symbols.
An overview of the frame structure can be found in Fig.
3. Synchronization is achieved in LTE with synchronization
sequences that occupy the 62 central subcarriers [5]. PSS and
SSS further carry the physical layer identity of the cell, the
cyclic prefix mode and also inform the user equipment (UE)
whether the cell uses frequency division duplex (FDD) or time
division duplex. The structure of PSS and SSS is also shown
in Fig. 3. Both synchronization signals are transmitted twice
per frame. For the PSS, the same sequence is transmitted each
time, while for the SSS, a different sequence appears in slot 1
and 11. Together, they indicate the physical layer cell identity.
The detection of the PSS is achieved by computing the
criterion according to (13) at each time position over half
a frame. Three criteria must be computed with, one for
each different PSS corresponding to a possible cell identity
within the group. The three criteria are then compared to
a threshold. Each position for which one of the criteria is
greater than a given threshold is considered as a possible
detection of a primary synchronization sequence transmitted
by a base station. The detection threshold is select for a
multi-channel receiver
SMB
SMB
digitized
samples
PCMCIA
SMB
FI signal 4
control computer
PCMCIA
SMB
SMA
FI signal 3
SMA
SMA
RF signal 4
FI signal 2
SMA
SMA
RF signal 3
FI signal 1
SMA
RF signal 2
SMA
SMA
RF signal 1
multi-channel acquisition
board
hard
disk
LTE GUI
USB
USB
RS232
GPS
coordinates
frequency and gain command
Fig. 4.
LTE signal
processing
Block diagram of multi-antenna LTE sensing platform
chosen probability of detection (POD) according to a desired
probability of false alarm (PFA).
The detection of the SSS is done with the same algorithm
for the positions, where a potential PSS has been detected.
For each of these time positions, 168 different sequences have
to be tested to obtain the identity of the group. Each SSS is
tested at 8 different timing positions: 2 for the possible CP
modes, 2 for the possible duplex modes and 2 for the possible
position in the frame. Here, no threshold is applied and from
all calculations, the combination yielding the highest criterion
is chosen.
IV. Experimental validation of the LTE sensing platform
The sensing platform consists of three units as shown in
Fig. 4 and Fig. 6. Filtering and gain control are applied
to the signal in the multi-channel receiver unit, the multichannel acquisition board is used to convert the signals to
digital domain and a control computer handles processing and
evaluation of the digital samples.
The multi-antenna LTE sensing platform is validated with
lab tests by measuring sensitivity and co-channel interference
rejection with real LTE eNBs. A hardware array simulator
consisting of splitters, coupling modules and a set of cables
of particular lengths is employed to virtually create a multiantenna, mono-path propagation channel with two directions
of arrival.
LTE eNB #2
LTE eNB #1
variable
attenuator
Fig. 6.
Multi-antenna LTE sensing hardware
Measurements of the interference rejection are performed
with the setup depicted in Fig. 5. The first input of the
hardware array simulator is connected to the LTE evolved
NodeB (eNB) producing the useful signal, while the signal
of an interfering eNB is provided on the second channel.
The interfering base station operates in 20 MHz mode with
fixed transmit power, while the to-be-detected base station
transmits a 5 MHz signal and the received signal strength is
gradually lowered with a variable attenuator. The detection
algorithm uses 8 frames for averaging the detection criterion,
the detection threshold is chosen for a PFA for 1% and the
target POD is set to 80%.
The results of the experiments are in Table II. First of all, it
hardware array simulator
sensitivity
level of 1st
eNB
multi-antenna LTE
sensing platform
coupling module
Fig. 5. Experimentation setup with hardware array simulator providing a
fixed angle of arrival
rejection
capacity
2nd eNB
1 antenna
4 antennas
multi-antenna gain
-92 dBm
-124 dBm
32 dB
11 dB
43 dB
32 dB
of
TABLE II
Interference rejection test results
V. Cognitive Radio Experimentation Methodology
The experiment that lead to the proof of concept in this
work has been conducted in framework of cognitive radio
experimentation methodology developed in the CREW project
[6]. Particularly, it is part of the usage scenario ”Cognitive
Radio in Cellular Networks” [7]. CREW is a federation of
several testbeds among European partners and provides the
means to perform CR experiments by enabling access to
facilities and equipment on the one hand and supporting
experimenters with a framework and guidelines for successful
experiments on the other hand.
Among the functionality provided by the CREW federation
[8], three particular things have been of benefit for this experiment. Firstly, CREW provides several modes of operation
for the individual testbeds. For instance, before conducting the
multi-antenna sensing experiment, reference signal files have
been recorded in the LTE testbed and tested with the sensing
device before joining both in the same physical location.
During the experiment, the setup and the equipment have been
carefully monitored and recorded, as part of the benchmarking
methodology which is being suggested in order to enable
comparable and reproducible results. Finally, the parameters
of the experiment, as well as the results have been put into a
common data format that allows to sharing among individual
institutions.
PSS/SSS
level (dBm)
can be seen that the useful base station signal can be detected
with a SIR 32dB lower when using 4 antennas compare to
the 1 antenna, thanks to the interference rejection of the platform. When comparing this number to the simulation results
depicted in Fig. 7, actually a multi-antenna gain of 26 dB is to
be expected for the chosen combination of POD and PFA. The
6 dB difference between simulation and experimental results
can be explained with the fact that the base station hardware
is configured such that it emits the PSS and SSS at a 6 dB
higher level than the remaining resources. This becomes also
evident from the spectrum snapshot in Fig. 8.
channel BW
sampling BW
frequency (MHz)
Fig. 8.
Spectrum view of the measured signal
Thus, besides verifying the algorithm and giving a proof of
concept for the multi-antenna sensing algorithm and hardware
in a setup with real signals, this experiment is also an example
how the CREW facilities and methodologies can support CR
research.
More information about this experiment can be found on
the project website [6].
VI. Conclusions
This work presents the outcome of the experimental
validation of a multi-antenna LTE sensing algorithm and
device, which is relevant for cognitive radio use cases like
inter-cell interference analysis or spectrum sensing with
interference rejection capabilities. It demonstrates, how a
detection criterion that is derived from a multi-antenna
signal can be applied to the signal emitted from an LTE
base station in order to decide if a certain base station
signal is present or not. It shows that the practically
achieved 32dB gain from multi-antenna processing matches
simulation results. Lastly, the benefits of conducting this
proof of concept in the CREW testbed federation are outlined.
Acknowledgements
This work has been performed in the framework of the
ICT projects ICT-248454 QOSMOS and ICT-258301 CREW,
which are partly funded by the European Union.
References
1
26 dB
detection probability
0.8
0.6
1 antenna
2 antennas
3 antennas
4 antennas
0.4
0.2
0
-50
-40
-30
-20
SIR [dB]
-10
0
10
Fig. 7. Simulated probability of detection vs. SIR for an LTE signal with
PFA = 0.01, Nfr = 8 in a mono path channel without mobility and in presence
of an interfering signal
[1] Depierre, D.; Pipon, F.; Loubaton, P.; Chaufray, J.-M.; ”Comparative
performance of UMTS-FDD downlink Multi-channel demodulators with
their extension to space tiem transmit diversity”, IST Summit 2003
[2] 1. Mitola, J.; ”Cognitive Radio: An Integrated Agent Architecture for
Software Defined Radio”, Royal Institute of Technology, May 2000
[3] Qing Zhao; Sadler, B.M.; , ”A Survey of Dynamic Spectrum Access,”
Signal Processing Magazine, IEEE , vol.24, no.3, pp.79-89, May 2007
[4] Cabric, D.; Mishra, S.M.; Brodersen, R.W.; , ”Implementation issues in
spectrum sensing for cognitive radios,” Signals, Systems and Computers,
2004. Conference Record of the Thirty-Eighth Asilomar Conference on
, vol.1, no., pp. 772- 776 Vol.1, 7-10 Nov. 2004
[5] 3GPP TS 36.211 ”Physical Channels and Modulation”
[6] www.crew-project.eu
[7] CREW D2.1 ”Definition of Internal Usage Scenarios” available
at http://www.crew-project.eu/sites/default/files/deliverables/CREW D2.
1 TUD R PU 2011-01-31 final PRC.pdf
[8] CREW
D3.1
”Basic
Operational
Platform”
available
at
http://www.crew-project.eu/sites/default/files/CREW D3.1 TCF
R P 2011-09-30 final.pdf