Experimental Study of Spectrum Sensing based on Energy Detection and Network Cooperation

Experimental Study of Spectrum Sensing based on
Energy Detection and Network Cooperation
Danijela Cabric
Berkeley Wireless Research Center
2108 Allston Way #200
Berkeley, CA 94704
++ 1-510-666-3192
danijela@eecs.berkeley.edu
Artem Tkachenko
Berkeley, CA 94704
++ 1-510-666-3100
artemtk@eecs.berkeley.edu
Robert W. Brodersen
Berkeley, CA 94704
++ 1-510-666-3110
rb@eecs.berkeley.edu
ABSTRACT
Spectrum sensing has been identified as a key enabling
functionality to ensure that cognitive radios would not interfere
with primary users, by reliably detecting primary user signals.
Recent research studied spectrum sensing using energy detection
and network cooperation via modeling and simulations. However,
there is a lack of experimental study that shows the feasibility and
practical performance limits of this approach under real noise and
interference sources in wireless channels. In this work, we
implemented energy detector on a wireless testbed and measured
the required sensing time needed to achieve the desired
probability of detection and false alarm for modulated and
sinewave-pilot signals in low SNR regime. We measured the
minimum detectable signal levels set by the receiver noise
uncertainties. Our experimental study also measured the sensing
improvements achieved via network cooperation, identified the
robust threshold rule for hard decision combining and quantified
the effects of spatial separation between radios in indoor
environments.
1. INTRODUCTION
Recently, Cognitive Radios (CRs) have been proposed as a
possible solution to improve spectrum utilization via opportunistic
spectrum sharing. Cognitive radios are considered lower priority
or secondary users of spectrum allocated to a primary user. Their
fundamental requirement is to avoid interference to potential
primary users in their vicinity. Spectrum sensing has been
identified as a key enabling functionality to ensure that cognitive
radios would not interfere with primary users, by reliably
detecting primary user signals. In addition, reliable sensing
creates spectrum opportunities for capacity increase of cognitive
networks.
The first application of spectrum sensing is studied under IEEE
802.22 standard group [1] in order to enable secondary use of
UHF spectrum for a fixed wireless access. In addition, there is a
number of indoor applications where spectrum sensing would
increase spectrum efficiency and utilization. For example, it could
improve co-existence of WLANs (802.11) [2], or create new
spectrum opportunities for sensor and ad-hoc networks [3].
Regardless of application, sensing requirements are based on
primary user modulation type, power, frequency and temporal
parameters. For example, the actual primary signal used for
sensing could be the regular data transmission signal.
Alternatively, it could be a special permission or denial signal to
use the spectrum, in the form of a pilot or a beacon.
Spectrum sensing is often considered as a detection problem,
which has been extensively researched since early days of radar
[4]. However, the key challenge of spectrum sensing is the
detection of weak signals in noise with a very small probability of
miss detection, which requires better understanding of very low
SNR regimes [5]. In addition, spectrum sensing is a cross-layer
design problem in the context of communication networks.
Cognitive radio sensing performance can be improved by
enhancing radio RF front-end sensitivity, exploiting digital signal
processing gain, and using network cooperation where users share
their spectrum sensing measurements [6].
Our goal is to provide a comprehensive study, supported with
experimental data, that addresses the following issues in spectrum
sensing based on energy detection:
-Required sensing time needed to achieve the desired
probability of detection and false alarm.
-Limitations of the energy detector performance due to
presence of noise uncertainty and background interference.
-Performance improvements offered by network cooperation.
How does the performance scale with the number of radios? What
is the robust threshold rule? What is the effect of spatial
separation between cooperating radios?
The paper is organized as follows: Section 2 reviews the energy
detector model, derives its performance and addresses the
limitations. In section 3, we provide the experimental data that
verifies theoretical results and characterizes energy detector
performance in noise. In section 4, we discuss the cooperation
gains in fading channels. Section 5 presents the experimental data
for cooperation. Summary of the work and conclusions are
presented in Section 6.
2. ENERGY DETECTION
CHARACTERIZATION
2.1 Model
We consider the detection of a weak deterministic signal in
additive noise. The signal power is confined inside a priori
known bandwidth B around central frequency fc (Figure 1). We
assume that activity outside of this band is unknown. Two

deterministic types of signal are considered: sinewave tone (pilot)
and modulated signal with unknown data.
An optimal detector based on matched filter is not an option since
it would require the knowledge of the data for coherent
processing. Instead a suboptimal energy detector is adopted,
which can be applied to any signal type. Conventional energy
detector consists of a low pass filter to reject out of band noise
and adjacent signals, Nyquist sampling A/D converter, square-law
device and integrator (Figure 2.a).
Without loss of generality, we can consider a complex baseband
equivalent of the energy detector. The detection is the test of the
following two hypotheses:
H0: Y [n] = W[n] signal absent
H1: Y [n] = X[n] +W[n] signal present
n = 1,…, N; where N is observation interval (1)
The noise is assumed to be additive, white and Gaussian (AWGN)
with zero mean and variance σw
2
. In the absence of coherent
detection, the signal samples can also be modeled as Gaussian
random process with variance σx
2
. Note that over-sampling would
correlate noise samples and, in principle, the model could be
always reduced to (1).
A decision statistic for energy detector is:
∑=
N
nYT 2
])[( (2)
Note that for a given signal bandwidth B, a pre-filter matched to
the bandwidth of the signal needs to be applied. This
implementation is quite inflexible, particularly in the case of
narrowband signals and sinewaves. An alternative approach could
be devised by using a periodogram to estimate the spectrum via
squared magnitude of the FFT, as depicted in Figure 2.b). This
architecture also provides the flexibility to process wider
bandwidths and sense multiple signals simultaneously. As a
consequence, an arbitrary bandwidth of the modulated signal
could be processed by selecting corresponding frequency bins in
the periodogram.
In this architecture, we have two degrees of freedom to improve
the signal detection. The frequency resolution of the FFT
increases with the number of points K (equivalent to changing the
analog pre-filter), which effectively increases the sensing time. In
addition, increasing the number of averages N also improves the
estimate of the signal energy. In practice, it is common to choose
a fixed FFT size to meet the desired resolution with a moderate
complexity and low latency. Then, the number of spectral
averages becomes the parameter used to meet the detector
performance goal. We consider this approach in our experiments.
2.2 Performance
It is well known that under the common detection performance
criteria (most notably, the Neyman-Pearson criteria) likelihood
ratio yields the optimal hypothesis testing solution and
performance is measured by a resulting pair of detection and false
alarm probabilities (Pd, Pfa). Each pair is associated with the
particular threshold γ that tests the decision statistic:
T > γ decide signal present
T < γ decide signal absent
When the signal is absent, the decision statistic has a central chi-
square distribution with N degrees of freedom. When the signal is
present, the decision statistic has a non-central chi-square
distribution with the same number of degrees of freedom. Since
we are interested in the low SNR regime, the number of required
samples is large. If N >250 we can use the central limit theorem to
approximate the test statistic as Gaussian.
T~ Normal(Nσw
2
, 2Nσw
4
) under H0
T~ Normal(N(σw
2
+ σx
2
) , 2N(σw
2
+ σx
2
)2
) under H1
Then Pd and Pfa can be evaluated as:
)
2
(
4
2
w
w
fa
N
N
QP
σ
σγ −
= )
)(2
)(
(
222
22
xw
xw
d
N
N
QP
σσ
σσγ
+
+−
= (3)
Note that for the constant false alarm probability (CFAR), the
threshold γ can be set even without the knowledge of the signal
power. Then, for the fixed number of samples N, Pd can be
evaluated by substituting the threshold in (3). Each threshold
corresponds to a pair (Pfa, Pd), representing the receiver operating
curve (ROC).
If the number of samples used in sensing is not limited, an energy
detector can meet any desired Pd and Pfa simultaneously. The
minimum number of samples is a function of the signal to noise
ratio SNR= σx
2
/ σw
2
:
[ ]21111
)())()((2 ddfa PQSNRPQPQN −−−−
−−=
(4)
In the low SNR << 1 regime, number of samples required for the
detection, that meets specified Pd and Pfa, scales as O(1/SNR2
).
This inverse quadratic scaling is significantly inferior to the
optimum matched filter detector whose sensing time scales as
Figure 1. Spectrum Picture
a)
b)
Figure 2. a) Implementation with analog pre-filter and square-law
device b) implementation using periodogram: FFT magnitude
squared and averaging

O(1/SNR) [5]. The exception is the sinewave case, where the
optimum matched filter is the FFT with length equal to the
multiple of sinewave period. If the FFT length is not matched,
then spectral leakage occurs. Therefore, the implementation in
Figure 2.b) is partially coherent for sinewave sensing, and it is
expected that its sensing time scales better than O(1/SNR2
).
2.3 Limitations
Unfortunately, an increased sensing time is not the only
disadvantage of the energy detector. More importantly, there is a
minimum SNR below which signal cannot be detected, and when
the formula (4) no longer holds. This minimum SNR level is
referred to SNRwall [7]. In order to understand when the detection
becomes impossible we need to revisit our signal model. There,
we have made two very strong assumptions (that are typically
made in communications system analysis). First, we assumed that
noise is white, additive and Gaussian, with zero mean and known
variance. However, noise is an aggregation of various sources
including not only thermal noise at the receiver and underlined
circuits, but also interference due to nearby unintended emissions,
weak signals from transmitters very far away, etc. Second, we
assumed that noise variance is precisely known to the receiver, so
that the threshold can be set accordingly. However, this is
practically impossible as noise could vary over time due to
temperature change, ambient interference, filtering, etc. Even if
the receiver estimates it, there is a resulting estimation error due
to limited amount of time. Therefore, our model needs to
incorporate the measure of noise variance uncertainty.
How does the noise uncertainty affect detection of signals in low
SNR? Essentially, setting the threshold too high based on the
wrong noise variance, would never allow the signal to be
detected. If there is a x dB noise uncertainty, then the detection is
impossible below SNRwall=10log10[10(x/10)
-1]dB [7]. For example,
if there is a 0.5 dB uncertainty in the noise variance, then signal in
-21dB SNR cannot be detected using energy detector.
3. ENERGY DETECTOR EXPERIMENTAL
RESULTS
The goal of our experimental study was to evaluate and verify the
theoretical results on the performance and limitations of the
energy detector. In particular, we measured the achievable
probabilities of detection as a function of sensing time, and the
existence and position of the SNRwall. The need for experiments is
stressed by the inability to realistically model all noise source
encountered in the receiver and interference environment. In
addition, a comprehensive evaluation of Pd and Pfa requires
extensive Monte Carlo simulations. Therefore, the
implementation on a real-time testbed allows us to perform a large
set of experiments for various signal levels and receiver settings.
We also provide an example of hardware implementation with
report on computational complexity, area, and speed. To the best
of our knowledge, this is the first energy detector study that
incorporates hardware realization and real-time experiments.
3.1 Testbed Description
The testbed used in the experiments [8] is built around the
Berkeley Emulation Engine 2 (BEE2), a generic multi-purpose
FPGA based, emulation platform for computationally intensive
applications. The BEE2 consists of 5 Vertex-IIPro70 FPGAs, 1
for control and 4 for user applications. Control FPGA runs Linux
and a full IP protocol stack convenient for connection with
laptops and other network devices. Linux OS has been enhanced
to allow access to hardware registers and memory on the user
FPGA for real-time data access and control. We used BWRC-
developed automation tool to map our signal processing
algorithms design in Xilinx System Generator library to FPGA
configurations.
BEE2 can connect up to 18 front-end boards via 10 Gbit/s full
duplex Infiniband interfaces. By using optical transceivers
compatible with Infiniband connectors, optical cable can connect
front-end boards at distances up to 1/3 of a mile away from BEE2
in order to perform different scenario experiments and implement
network cooperation. In addition, the optical link provides good
analog signal isolation on the front-end side from the digital noise
sources created by BEE2.
The radio front-end system operates in 2.4 GHz ISM band over 85
MHz of bandwidth with programmable center frequency and
several gain control stages. Antennas used in the experiments are
single monopole rubberduck with 0 dBi omnidirectional pattern at
2.4 GHz. The analog/baseband board contains a 14-bit 128 MHz
D/A converters, 12-bit 64 MHz A/D converters, and 32 MHz
wide baseband filters. On board Virtex-IIPro20 is used to
implement radio control functions and provide optical transceiver
interface to BEE2 for sample processing.
For the transmitter, we used Agilent EE4438C ESG vector signal
generator. It is calibrated to output absolute signal levels for
arbitrary signals and modulation types. The transmitter was
interfaced to BEE2 via 100 Base-T Ethernet so that signal
parameters can be changed during the runtime of the experiments.
The test setup is presented in Figure 3.
3.2 Energy Detection Implementation
The energy detector is implemented using 1024 point FFT with a
fully parallel pipelined architecture for the fastest speed. Due to
A/D sampling at 64 MHz, this implementation has 62.5 kHz FFT
bin resolution. Each block of FFT outputs is averaged using an
accumulator with programmable number of averages. The result
of the computation is stored in the memory block RAM. The
Figure 3. Testbed: Signal generator, Radio board, BEE2 and optical cable

software running on the BEE2 control FPGA sets all sensing
parameters and then loads the processed data from the block
RAMs. The design runs at 100 MHz speed, while the signal
samples from the radio are fed at 64MHz. Thus, there is an
insignificant latency in the signal processing. For example,
executing 1000 experiments with 3200 spectral averages (51.2
ms) of 1024 FFT takes less than 60 seconds. Hardware estimate of
this design is: 15,416 FPGA logic slices, 140 18x18 multipliers,
and 106 block RAMs.
3.3 Experimental Setup
To measure the performance under AWGN we connect signal
generator to the RF board antenna input via SMA cable. The radio
is put inside the RF shield, thus the only noise sources come from
the radio circuitry. Prior to all experiments, we calibrated the
noise level of the radio receiver, and the measured level is -103
dBm in a 62.5 kHz FFT bin. In order to accommodate a wide
dynamic range signal and keep A/D resolution high, we set the
total receiver gain to medium level of 50dB.
We tested two types of signals: sinewave carrier at 2.493GHz,
and 4 MHz wide QPSK signal centered at the same carrier. For
sinewave carrier we swept signal levels from -110 dBm to -128
dBm, which is equivalent to -7 to -25 dB of the receiver SNR. For
4 MHz QPSK signal, we tested levels from -98 dBm to -110 dBm,
which is equivalent to -13 to -25 dB of the receiver SNR. In order
to accurately estimate the Pd and Pfa we repeated each detection
measurement 1000 times. For each signal level, we collected two
sets of energy detector outputs: one in the presence, and the other
in the absence of the signal generator output signal. From “no
input signal” data, we estimated the detection threshold to meet
the specified probability of false alarm. Then, we applied the
threshold to the data where signal was present and computed the
probability of detection.
3.4 Results
First, we measure how the probability of detection scales as the
sensing time increases. For all measurements, we set the
probability of false alarm to 5%. Figure 4.a) shows the achievable
probabilities of detection for sinewave carrier when number of
averages increases from 200 (3.2 ms) to 52,000 (0.83 s). If we set
the Pd to a reasonable limit of 0.8 then within 200 ms energy
detector can detect up to -122 dBm. Figure 4.b) shows the
performance of the detector for the QPSK signal. Since the energy
of the QPSK signal is spread over a wider bandwidth, it becomes
harder to detect weak signals in additive noise. The target Pd=0.8
can be achieved for signals greater than -104 dBm within 170 ms.
Figure 4 also shows that when the signal becomes too weak,
increasing the number of averages does not improve the detection.
This result is expected and is explained by the SNRwall existence.
In Figure 5, we plot the required sensing time vs. input signal
levels in order to meet Pd and Pfa requirements simultaneously.
We set the Pfa=0.05 and Pd=0.6. In the case of sinewave sensing
we observe the scaling law N~1/SNR1.5
.This is expected since our
implementation has a partial coherent processing gain for
sinewave detection. However, beyond -124 dBm signal level the
slope becomes increasingly steep. Signals below -128 dBm
cannot be detected, resulting in the SNRwall=-25dB. In the case of
QPSK detection, we observe a non-coherent detection scaling law
of N~1/SNR2
consistent with the theoretical prediction. The limit
in QPSK detection happens at -110dBm (SNRwall=-25dBm). From
the theoretical analysis, we know that SNRwall=-25dB corresponds
to less than 0.5 dB of noise uncertainty.
4. COOPERATIVE SENSING
Up to this point we have considered spectrum sensing performed
by a single radio in AWGN-like channels. In fading channels,
however, single radio sensing requirements are set by the worst
case channel conditions introduced by multipath, shadowing and
local interference. These conditions could easily result in SNR
regimes below the SNRwall, where the detection will not be
possible. However, due to variability of signal strength at various
locations, this worst case condition could be avoided if multiple
radios share their individual sensing measurements via network
cooperation [9],[10],[11].
a) Sinewave sensing
b) QPSK sensing
Figure 4. Probability of detection vs. sensing time for Pfa=5%

4.1 Cooperation Gains
Under independent fading conditions, which is often assumed for
multipath if radios are more than λ/2 apart, cooperation can be
studied as a diversity gain in multiple antenna channels. Due to a
small overhead in the protocol, we consider a hard decision
combining, where each radio sends its local decision to a
centralized location (0 signal is absent, 1 signal is present), and
the decisions is made via OR operation. It has been shown [9] that
if n radios combine independent measurements, then probability
of detection of the system QD monotonically increases as QD=1-
(1-Pd)n
. In addition, the probability of false alarm for the system
QF also monotonically increases as QF=1-(1- Pfa)n
.
However, fading could be caused by shadowing that exhibits high
correlation if two radios are blocked by the same obstacle.
Commonly, a shadowing correlation is described by the
coefficient ρ and modeled as an exponential function of distance:
ρ=e-ad
. Measurements of the shadowing in indoor environments
[12] show that the correlation coefficient is independent of
wavelength over a frequency octave, but it is dependent on the
topography. It was estimated [12] that 90% correlation distance is
typically 1m, 50 % is around 2m, and slowly decays to 30% over
8m. Therefore, in the limited area, increasing the number of
radios introduces the correlation, which in effect limits the
cooperation gain [9], [11]. In our experimental study, we
investigate how the cooperation gain scales with the number of
radios and their spatial separation in typical indoor environments.
4.2 Threshold Rules
Recall the single radio analysis where we identified the SNRwall
due to noise uncertainty and its impact on the detection threshold.
Now, in the cooperation case, radios could use two different types
of threshold rules in the local decision process: 1) a
predetermined (fixed) threshold set by the centralized processor
or 2) an independently estimated threshold based on the local
noise and interference measurements. In the case of stationary
environments with all radios being identical, these two rules
would result in the same system performance. However, due to
the presence of ambient interference caused by primary or
cognitive radios in the vicinity, and local noise, temperature, and
circuit variability, each radio sees different aggregate noise and
interference. This observation suggests that a fixed threshold
might be suboptimal, and that in practical situations the estimated
threshold would provide robustness and better gains. Through
experiments, we analyze benefits of noise and interference
estimation, and the gap between the two threshold rules.
5. EXPERIMENTAL RESULTS FOR
COOPERATION
5.1 Experimental Setup
The experiments were conducted inside the Berkeley Wireless
Research Center. The floor plan of the center is shown in Figure
6. The figure also shows 54 locations on a 2m by 2m grid, that
covers a cubicle area, library and conference room, where all
wireless measurements were taken. In all experiments, the
a) Sinewave sensing
b) QPSK sensing
Figure 5. Required sensing time vs. signal input level for fixed Pd
and Pfa
Figure 6. BWRC floor plan with transmitter and receiver locations

transmitter was located inside the lab. Therefore, the signal path
between the transmitter and all receiver positions included
propagation through either concrete or wooden walls, supporting
beams, medium and large size metal cabinets, and general office
furniture. The area covers a balanced variation of obstacles which
are typical for indoor non-line-of-sight environments.
Due to operation in the unlicensed ISM band, outside interference
had to be considered. All 802.11 b/g, Bluetooth, and ZigBee
equipment was shut down during the experimentation, in order to
minimize potential interference. For the sinewave, the signal
generator transmitted a -40 dBm signal at 2.485GHz. For the 4
MHz wide QPSK signal, the signal generator transmitted a -30
dBm signal in 2.483GHz – 2.487GHz band, centered at
2.485GHz. The bandwidth ratio of QPSK signal to sinewave is
approximately 10*1og10(4 MHz/62.5kHz)=18 dB, thus a 10 dB
difference in transmit power favors the sinewave case in terms of
the receiver SNR. It was expected that sinewave performance
would be more affected by multipath, thus 8 dB power gain was
added.
For each sensing location, data was collected for three different
transmitter configurations: idle spectrum i.e. no signal, sinewave
signal and QPSK signal. The idle spectrum was sensed in order to
be able to compare two different threshold rules described in the
previous sections. For each location and data type, spectrum was
sensed 200 consecutive times using 3200 averages (51.2 ms) in
the periodogram.
5.2 Measurement Results
First, we analyze the cooperation gain as function of the number
of cooperating radios. Figures 7 a) and b) show the system ROCs
for sinewave and wideband QPSK signals, respectively. For a
given probability of false alarm for the cooperating system QF, an
estimated threshold was computed for each location based on the
idle spectrum data. Then, these thresholds were used to compute
the probability of detection for each location. By applying the OR
function to decisions of n radios and averaging across all possible
combination of n radios among 54 locations, we obtain QD.
For the sinewave signal, the single radio sensing is limited by
multipath fading, thus significant improvement is achieved even
with 2 cooperative radios. Given 10% probability of false alarm,
an 18% improvement is observed through cooperation of two
radios, then 9% for three, and it saturates to 4% and 3% for four
and five cooperating radios, respectively. Overall, going from 1 to
5 cooperative radios detection improves from 63% to 97%. Note
that if radios would experience the independent multipaths then
the cooperation would result in QD=1-(1-0.63)5
=99%. In the case
of wideband QPSK signal, the probability of detection is even
better, though the average SNR is 8dB lower. With 5 cooperative
radios, QD for QPSK reaches 99%. This improvement in the
QPSK sensing is due to frequency diversity gain, which makes
the wideband signal less sensitive to deep fades.
Next, we compare experimental results for two different threshold
rules applied to the same set of measurement data. For the fixed
threshold rule, the same threshold is applied for all possible
combinations of cooperating radios. As expected, Figure 8 shows
a) Sinewave sensing
b) QPSK sensing
Figure 7. Cooperative gains vs. number of cooperative radios Figure 8. Cooperative gains vs. threshold rule

that the performance of the fixed threshold rule is significantly
lower than that of the estimated threshold rule. Even for the
single radio sensing under variable noise and interference, it is
essential to apply the location and time relevant threshold
obtained via estimation. The cooperation gains are still present,
but are significantly reduced by suboptimal threshold rule. The
gap between the two rules varies form 15% to 25%. Thus, even a
moderate QD=90 % and QF=10% can never be met using the
fixed threshold. The implications of this result imply that robust
sensing must involve frequent receiver noise calibration and
accurate interference estimation. In turn, this might require
additional sensing time.
Lastly, we analyze the effect of spatial correlation on the
probability of detection for two cooperating radios. Figure 9
presents experimental results for sinewave and 4 MHz QPSK
signals for separation distances from 2 m to 8 m. Measurements
show that the probability of detection monotonically increases as
the separation between two cooperating radios increases. The
improvement of up to 7% in QD is obtained once the cooperation
distances extend to 8 m. This result can be explained by
measurements in [12] showing that correlation coefficient of
50% for distances of 2 m is still high to provide independency
needed for cooperation gain. In [9], simulation study has shown
that cooperation gain increases QD by 2-3% as the correlation
coefficient drops from 50% to 10%. Our measurements confirm
that it is beneficial to increase the radio separation so that
correlation coefficient is less than 10%, which for this particular
environment happens at 8 m distances.
6. CONCLUSIONS
In this paper we investigated the feasibility and performance of
the energy detector for the spectrum sensing of narrowband
pilots and wideband primary user signals. Our study includes a
theoretical background and experimental results for two
applications: 1) single radio sensing performed on the physical
layer and 2) cooperative sensing performed by a network of
cognitive radios in an indoor environment. We derived and
measured the required sensing time, achievable probability of
detections and false alarm, and minimum detectable signal levels
in AWGN and fading channels. Experiments are performed in
the 2.4 GHz ISM band using a radio testbed whose front-end and
baseband circuitry exhibit true noise, gain and filtering variability.
In AWGN-like channels, it was found that the presence of radio
uncertainties sets practical limits on minimum detectable signal
levels which cannot be further improved by signal processing.
In fading channels, single radio sensing is limited by the worst
case channel conditions introduced by multipath and shadowing.
We deployed multiple radios in a typical office indoor
environment and measured sensing improvements achieved via
cooperation. Our cooperation study also identified the robust
threshold rule for hard decision combining. In addition, we
quantified the effects of spatial separation between cooperating
radios in the typical indoor environments.
7. ACKNOWLEDGMENTS
This work was supported by MARCO Contract CMU 2001-CT-
888, and the industrial members of BWRC.
8. REFERENCES
[1] C. Cordeiro, K. Challapali, K. Birru, S. Shankar, “IEEE
802.22: the first worldwide wireless standard based on
cognitive radios”, In proc. of DySPAN’05, November 2005.
[2] D. Cabric, S.M. Mishra, D. Willkomm, R.W. Brodersen, A.
Wolisz, “A Cognitive Radio Approach for Usage of Virtual
Unlicensed Spectrum”, 14th
IST Mobile and Wireless
Communications Summit, June 2005.
[3] B. Wild, K. Ramchandran, “Detecting Primary Receivers for
Cognitive Radio Applications”, In proc. of DySPAN’05,
November 2005.
a) Sinewave sensing
b) QPSK sensing
Figure 9. Cooperation gains vs. spatial separation

[4] H.Urkowitz,“Energy Detection of Unknown Deterministic
Signal”, In proc. of the IEEE, April 1967.
[5] A.Sahai, N.Hoven, R. Tandra. “Some Fundamental Limits
on Cognitive Radio”, In Allerton Conference on
Communications, Control and Computing, October 2004.
[6] D.Cabric. S.M. Mishra. R.W. Brodersen, “Implementation
Issues in Spectrum Sensing”, In Asilomar Conference on
Signal, Systems and Computers, November 2004.
[7] R. Tandra, A. Sahai. “Fundamental Limits on Detection in
Low SNR”, In proc. of the WirelessComm05 Symposium on
Signal Processing, June 2005.
[8] S.M.Mishra, D. Cabric, C. Chang, D. Willkomm, B. van
Schewick, A. Wolisz, R. Brodersen, “A Real Time Cognitive
Radio Testbed for Physical and Link Layer Experiments” In
proc. of DySPAN’05, November 2005
[9] A. Ghasemi, E. S. Sousa, “Collaborative Spectrum Sensing
for Opportunistic Access in Fading Environments”, In proc.
of DySPAN’05, November 2005.
[10] E. Visotsky, S. Kuffner, R. Peterson. “On Collaborative
Detection of TV Transmissions in Support of Dynamic
Spectrum Sharing”, In proc. of DySPAN’05, November
2005.
[11] S.M.Mishra, A.Sahai, R.W.Brodersen,”Cooperative Sensing
among Cognitive Radios”, In proc. of International
Conference on Communications, ICC’06, June 2006.
[12] J.C. Liberti, T.S. Rappaport, “Statistics of shadowing in
indoor radio channels at 900 and 1900 MHz”, In proc. of
Military Communications Conference, MILCOM’92,
October 1992.

Experimental Study of Spectrum Sensing based on Energy Detection and Network Cooperation

More Related Content

What's hot (18)

Similar to Experimental Study of Spectrum Sensing based on Energy Detection and Network Cooperation (20)

Recently uploaded (20)

Experimental Study of Spectrum Sensing based on Energy Detection and Network Cooperation