This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts

for publication in the IEEE ICC 2011 proceedings

Spectrum Sensing via Energy Detector in Low SNR

Saman Atapattu, Chintha Tellambura, and Hai Jiang
Department of Electrical and Computer Engineering, University of Alberta, Canada
Email: {atapattu, chintha, hai.jiang}@ece.ualberta.ca

Abstract—As required in the IEEE 802.22 proposal, spectrum processes, the decision statistic is modeled with a Gaussian
sensing techniques should be capable enough to sense the primary distribution using the central limit theorem (CLT) [11]–[13].
signal with very low receiver sensitivity such as at -116 dBm. In The Gaussian model is popular in the parameter optimiza-
this paper, the detection performance of an energy detector used
for spectrum sensing in cognitive radio networks is investigated tion problems, e.g., optimizing the operating threshold or the
under such very low signal-to-noise ratio (SNR) levels. The power allocation so as to achieve the maximal throughput or
analysis focuses on the derivation of a closed-form expression for minimal error rate. This model often gives a more convenient
the average missed-detection probability over Rayleigh fading cost function which may result in a convex optimization
and Nakagami-m fading channels. Subsequently, the detection problem. However, based on the Gaussian model, the analysis
threshold is optimized for minimizing the total error rate. The
analysis is validated by numerical and simulation results. The of the average detection performance of an energy detector
sensing requirements defined in IEEE 802.22 are also discussed over different fading scenarios is not available in the open
with numerical examples. literature because of the involved mathematical complexity.
Index Terms—Cognitive radio, energy detection, spectrum The existing analytical results are limited to the additive white
sensing, threshold selection. Gaussian noise (AWGN) channel, and performance over other
fading scenarios are obtained only by simulations.
I. I NTRODUCTION In this paper, we derive the average missed-detection proba-
bility of an energy detector in low signal-to-noise ratio (SNR)
One of the most challenging tasks in cognitive radio net- region over Rayleigh fading and Nakagami-m fading channels.
works is spectrum sensing. In the IEEE 802.22 wireless re- The low SNR assumption is fairly reasonable because, as
gional area networks (WRAN) proposal, no specific spectrum in IEEE 802.22 WRAN, the spectrum sensing technique
sensing technique is given. So designers have freedom to select should be able to detect the primary signal with the missed-
any spectrum sensing technique to meet the specified sensing detection and the false alarm probabilities less than 0.1 and
requirements [1]. Among the available spectrum sensing tech- the receiver sensitivity being -116 dBm [1], [14], [15]. More
niques such as matched filter, cyclostationary feature detection importantly, we determine the optimal detection threshold of
and eigenvalue detection, energy detection has gained renewed the energy detector to minimize the total error rate. Based on
interests in recent research efforts due to its low complexity. the analytical results, some numerical examples are given to
The conventional energy detector measures the energy associ- meet the IEEE 802.22 WRAN requirements.
ated with the received signal over a specified time period and The rest of this paper is organized as follows. Section II
a bandwidth. The decision of an energy detector is feasible briefly discusses energy detection and its low SNR model.
even when little prior knowledge of the transmitted signal Section III gives the average missed-detection probability.
is available. The decision statistic of an energy detector is Sections IV is devoted to the analysis of the optimal detection
a measure of the received signal energy after proper filtering, threshold. Section V presents numerical and simulation results,
sampling, squaring and integration. followed by concluding remarks in Section VI.
Assuming a deterministic signal is transmitted over a flat
band-limited Gaussian noise channel, a basic mathematical II. E NERGY D ETECTION AND L OW SNR A PPROXIMATION
model of the decision statistic is given in [2] in order The spectrum sensing in cognitive radio networks follows
to calculate the detection probability (Pd ) and false alarm a binary hypothesis testing problem: hypothesis H0 (signal
probability (Pf ). Subsequently, the performance of an energy absent) and hypothesis H1 (signal present). The received
detector in terms of the average detection probability, the signal for the binary hypothesis can be given as
receiver operating characteristic (ROC), and the area under 
w(t) : H0
the ROC curve (AUC) over different fading channels, diversity y(t) =
hx(t) + w(t) : H1
techniques and cooperative relay networks has been analyzed
in [3]–[10]. In spectrum sensing of cognitive radio networks, where x(t) is the transmitted signal, h is the wireless channel
the secondary user has no a priori knowledge of the primary gain, and w(t) is the additive white Gaussian noise (AWGN)
signal. The information bearing signal can have different which is assumed to be a circularly symmetric complex Gaus-
possible waveforms with random data sequences. Therefore, sian (CSCG) random variable with mean zero and variance
2
it is appropriate to treat the received signal samples as a σw . The conventional analog energy detector consists of a
random process. When both signal and noise follow Gaussian pre-filter followed by a square-law device and a finite time

978-1-61284-233-2/11/$26.00 ©2011 IEEE

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2011 proceedings

integrator. The output of the integrator is called the decision 1

statistic. It may be proportionalto the received signal energy
N 2
and can be given as Λ(y) = n=1 |y(n)| where N is the
0.9
− 20 dB
number of samples. When the signal has an unknown form, it 0.8
− 15 dB
is appropriate to treat the signal samples as a random process. 0.7 − 10 dB
Thus, the sample of the transmitted signal x(n) follows an

CDF (F (y,H ))
0.6

1
independent and identically distributed (i.i.d.) random process
with mean zero and variance σs2 . Then, the received SNR at

Λ
0.5

|h|2 σ 2
the detector is γ = σ2 s for the given channel h. When the 0.4
w
sample number N is large enough, using CLT, the probability 0.3

density function (PDF) of Λ(y) under H0 , fΛ (y, H0 ), is 0.2

2 4
a normal distribution with N σw mean and N σw variance. 0.1 Exact
Similarly, the PDF of Λ(y) under H1 , fΛ (y, H1 ), is a normal Low SNR approximation
2 4
distribution with N σw (1+γ) mean and N σw (1+2γ) variance 0
1800 1900 2000 2100 2200 2300 2400
y
for a complex-valued phase-shift keying (PSK) modulated
signal. The false alarm probability, Pf , and the detection
Fig. 1. CDF of the decision statistic under hypothesis H1 .
probability for given h, Pd|h (γ), are given as
  
λ √
Pf = Q 2
− 1 N (1)
N σw fading and Nakagami-m fading channels, which can model
   a variety of fading effects. If the signal amplitude follows
λ N a Nakagami-m distribution, the SNR distribution is given as
Pd|h (γ) = Q −1−γ , (2) m
N σw 2 (1 + 2γ) (m
γ̄ )
m
m−1 − γ̄ x
fγ N ak (x) = Γ(m) x e [16] where m is the fading
respectively, where λ is the threshold and Q(·) is the standard parameter, and it follows Rayleigh fading when m = 1. With
Q-function. the aid of the identity Erfc(−x) = 2 − Erfc(x), and after
Since IEEE 802.22 WRAN standard is interested in the some algebraic manipulations, the average missed-detection
probability over Nakagami-m fading channel, Pmd N ak , can be
spectrum sensing in very low SNR region (<-20 dB), we
consider a low SNR energy detection model. Under the low given as
SNR assumption (i.e., σs2 ≈ σw 2
), the signal has a little impact m
m  
∞
on the variance of the decision statistic under H1 . Therefore, γ̄ m−1 − mx N N σ 2−λ
2
N ak
Pmd = x γ̄ e Erfc x +√ dx. (4)
PDF fΛ (y, H1 ) is Gaussian distributed with N σw (1 + γ) 2Γ(m) 0 2 2N σ 2
4
mean and N σw variance. Fig. 1 shows cumulative distribution
function (CDF) of the exact and its low SNR approximation In the following, we define an integral expression, I(n, p, a, b),
for three different low SNR values when σw = 1 and as [17, eq. (2.8.9.1)]
N = 2000. The approximated and the exact CDFs are very ∞
close. Therefore, for low SNR, Pf and Pd|h (γ) can be given I(n, p, a, b)  xn e−px Erfc(ax + b)dx
0
in alternative forms as  p2 +4pab

 2
 4a2 Erfc(b+ 2a
p
)
1 λ − N σw ∂n
Erfc(b)−e
Pf = Erfc √ p
2 2N σw2
= (−1) n
 2
 ∂pn
1 λ − N σw (1 + γ)
Pd|h (γ) = Erfc √ , (3) n
2 2N σw 2 where n is a positive integer, Re[p]> 0, c > 0, and ∂p ∂
n [·] is

the nth order partial derivative with respect to p. For integer

respectively, where Erfc(·) is the complementary error function N ak given in (4) can be evaluated as
∞ 2 m, Pmd
which is defined as Erfc(z) = √2π z e−t dt and we have
m  
Q(z) = 12 Erfc √z . m
N N σ2 − λ 
2 γ̄ 
N ak =
Pmd I m − 1, p, , √  . (5)
III. AVERAGE M ISSED -D ETECTION P ROBABILITY 2Γ(m) 2 2N σ 2 
p= m
γ̄
The instantaneous detection probability given in (3) is
When m = 1, the average missed-detection probability over
equivalent to the average detection probability over AWGN Ray
channel when γ is replaced by the average SNR γ̄. Here, Rayleigh fading channel, Pmd , is
√
we are interested in the missed-detection probability (as it is    1 +4 N

2
√σ −λ N

one requirement in IEEE 802.22 WRAN), Pmd|h (γ), which 1 N σ2 − λ γ̄ 2 γ̄ 2N σ 2 2

Ray
Pmd = Erfc √ −e 2N

is given as Pmd|h (γ) = 1 − Pd|h (γ). When the SNR dis- 2 2N σ 2

  (6)
tribution is fγ (x), the average missed-detection probability, N σ2 − λ 1
∞ × Erfc √ + √ .
Pmd , is Pmd = 0 Pmd|h (x)fγ (x)dx. We consider Rayleigh 2N σ 2 γ̄ 2N
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2011 proceedings

When a random signal is present, the results in (5) and (6) and the observation made in the AWGN channel, we can
are novel closed-form expressions for the average missed- find an approximated optimal value. When γ̄  1, the right
detection probability. hand side of (9) approaches a very small value. We take
2
1 λ−N σw 2
IV. T HRESHOLD S ELECTION α = √2N γ̄
− √
2
2N σw
. When γ̄  1 and λ → N σw , the
left hand side of (9) also approaches a very small value due
The threshold, λ, which varies form 0 to ∞ is a common 2
to lim eα Erfc(α) → 0. Thus, we can say that the equation
parameter for the false alarm, the detection and the missed- α→0
2
detection probabilities which are denoted as Pf (λ), Pd (λ) (9) is satisfied for λ around N σw . Further, it can be shown
2
Pe (λ)
and Pmd (λ), respectively. The common practice of setting the that the second order derivative of Pe (λ), ∂ ∂λ 2 > 0 when
threshold is based on the false alarm probability. For given N , 2 Ray
λ ≈ N σw . Therefore, Pe = (Pf + Pmd ) has a minimum
σw , and the constant false alarm probability (CFAP=P̄f ), the 2
√ √ √ around λ ≈ N σw . This observation is also concluded with
selected threshold is λ = 2Erfc−1 (2P̄f ) + N 2
N σw . the numerical results given in Section V.
Although achieving a high Pd (λ) while keeping Pf (λ) low Note that the optimal threshold selection with a Nakagami-
is preferable (e.g., in IEEE 802.22 WRAN recommendations, m fading channel is analytically complicated because Pe =
P̄f ≤ 0.1 and P̄d ≥ 0.9), such a threshold selection is not (Pf +Pmd N ak ) has highly non-linear behavior. Since Nakagami-

always possible in practice. Therefore, the threshold selection m fading (when 1 < m < ∞) is varying between the
can be viewed as an optimization problem. Some research Rayleigh fading and the Gaussian fading, we can also claim
2
work has been done for this problem based on different that the optimal threshold is around λ ≈ N σw . However, the
objectives [11], [18]–[23]. The minimization of the total exact solution can be obtained numerically using mathematical
error rate which is defined as Pe (λ)  Pf (λ) + Pmd (λ) software packages such as MATHEMATICA and MATLAB.
is a possible way of selecting λ [18]. We consider different It is possible that we cannot achieve the recommended error
fading scenarios in the following subsections, and the optimal rate requirements (P f ≤ 0.1 and P md ≤ 0.1) even at the
threshold, λ∗ , is derived such that the total error is minimized, optimal threshold value. One possible way of achieving the
i.e., λ∗ = arg min Pe (λ). requirements is by increasing the number of samples N . Since
λ
N ≈ τ fs where τ is the sensing time and fs is the sampling
A. AWGN Channel frequency, the sensing time also increases when N increases.
For an AWGN fading channel, the optimal threshold is given However, there is a limitation for the allowable sensing time,
using (1) and (2) as i.e., τ ≤ 2 seconds in accordance with IEEE 802.22 WRAN
  2
 [1], [15]. This is a main drawback in spectrum sensing with
∗ 1 λ − N σw
λ = arg min 1+ Erfc √ energy detection in low SNR. This issue is discussed with
λ 2 2N σw2
  numerical examples in the following section.
2
1 λ − N σw (1 + γ)
− Erfc  . V. N UMERICAL /S IMULATION R ESULTS AND D ISCUSSION
2 2N (1 + 2γ)σw 2

Therefore, the optimal threshold for any SNR value can be This section provides numerical and simulation results. We
derived as (see the Appendix) defined the normalized threshold as λ̂  N λ
, i.e., the threshold
2
   
is normalized by the number of samples. The noise variance
2
N σ 2
(1 + 2γ) ln(1 + 2γ) is set to σw = 1, unless specified otherwise. We denote Pe∗ =
λ∗ = w
1 + 1+ 2γ 1+ . (7) Pe (λ ), Pf = Pf (λ∗ ), and Pmd
∗ ∗ ∗
= Pmd (λ∗ ).
2 N γ2
One of the main contributions of this research is to derive
In low SNR, i.e., γ  1, thus 1+2γ ≈ 1, the optimal threshold a closed-form expression for the average missed-detection
can be well-approximated as probability over Nakagami-m fading channel in low SNR.
2
N σw  Fig. 2 shows ROC curves (i.e., Pd versus Pf ) for three
λ∗ ≈ 1 + 1 + 2γ ≈ N σw 2
. (8) different fading scenarios such as AWGN, Nakagami-4 and
2
B. Rayleigh Channel Rayleigh fading channels. The numerical results which are
based on expressions (5) and (6) are represented by curves,
For a Rayleigh fading channel, the total error is Pe = (Pf + while simulation results are represented by discrete marks. The
Ray Ray
Pmd ) where Pmd is given in (6). When ∂P∂λ = 0, it can be
e
ROC curves are plotted by varying λ̂ from 0.95 to 1.50 for the
1
simplified as average SNR γ̄=-20 dB and γ̄=-15 dB when N = 2000. The
 2  
λ−N σw2
2 numerical results closely match with the simulation results
√ 1 −√ 2 1 λ − N σw 2N
e 2N γ̄ 2N σw
Erfc √ −√ = γ̄. (9) for all three fading scenarios at -15 dB and -20 dB, which
2N γ̄ 2N σw2 π confirms the accuracy of the approximation. The multipath
It is complicated to derive the exact solution for λ∗ with fading parameter m has a negligible impact on the energy
this non-linear equation. The solution can be obtained nu- detection at -20 dB, because the faded replicas of different
merically. However, with the assumption of very low SNR
2 This is the threshold
N if the decision statistic is selected as Λ(y) =
1 2
n=1 |y(n)| .
1 Due to space limitation, the detailed derivation is omitted. N
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2011 proceedings

1
P*e at −15 dB
0.9
P* at −15 dB
−15 dB f
0.8 P* at −15 dB
md
P*e at −20 dB
0.7
P* at −20 dB
f
0.6
P* at −20 dB

Error rates
md
Pd

0.5
−20 dB
0.4

0.3

0.2
AWGN
0.1 Nakagami − 4
Rayleigh
−1
0 10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 3 4 5
10 10 10
P N
f

Fig. 2. ROC curves of an energy detector over fading channels. Fig. 4. Minimum error rates versus number of samples in AWGN channel.

1
P* at −15 dB
e
P*f at −15 dB
0.9
P*md at −15 dB

P* at −20 dB
− 20 dB e
0.8
P*f at −20 dB
*
Error rates Pmd at −20 dB
Pe

0.7

0.6
− 15 dB

0.5
AWGN
Nakagami−5
Rayleigh
−1
0.4 10
0.65 0.7 0.75 0.8 0.85 3 4 5 6
10 10 10 10
Normalized Threshold
N

Fig. 3. Total error rate versus normalized threshold. Fig. 5. Minimum error rates versus number of samples in Rayleigh channel.

multipaths have no significant contribution to increase the WRAN expects Pf ≤ 0.1 and Pmd ≤ 0.1, and channel
effective SNR at very low SNR. detection time (CDT) τ ≤ 2 seconds under any detection
We show that the total error rate Pe = Pf + Pmd can be technique. Since N ≈ τ fs where fs is the sampling frequency
minimized at the optimal threshold value λ∗ . The total error which may depend on the sampling rate of the analog to digital
rate versus normalized threshold for AWGN, Nakagami-5 and converter and FFT (fast Fourier transform) bin resolution, we
Rayleigh fading channels is shown in Fig. 3 at γ̄ = −15 dB cannot increase N beyond τ fs . We take fs = 62.5 kHz which
and γ̄ = −20 dB when N = 2000. It shows that λ̂∗ ≈ 0.755 is a typical FFT bin resolution of an experimental energy
at γ̄= -20 dB, and λ̂∗ ≈ 0.760 at γ̄= -15 dB when σw 2
= 0.75 detection implementation [24]. We consider the minimum
for all three different fading scenarios. As in expression (8), error rate requirements as Pf ≤ 0.1 and Pmd ≤ 0.1. Fig.
λ∗
the optimal normalized threshold is given as λ̂∗ = N ≈ 4 and Fig. 5 show minimum error rates (Pe∗ , Pf∗ , and Pmd∗
)
2  √ 
σw 2
2 1 + 1 + 2γ ≈ σ w for AWGN, and it is also valid for versus N for AWGN and Rayleigh channels, respectively. For
Rayleigh and Nakagami-m fading channels. So the analytical -15 dB, the requirements can be achieved when N ≥ 7000
results in Section IV is confirmed. Further, we can see that one with Pe∗ ≈ 0.193 in AWGN channel. If we neglect other
possible way of minimizing the total error rate is by increasing processing delays, the minimum CDT is 0.112 seconds. For
the average SNR, e.g., Pe∗ can be reduced from 0.82 to 0.48 -20 dB, the requirements can be achieved when N ≥ 67000
by increasing SNR from -20 dB to -15 dB in AWGN channel. with Pe∗ ≈ 0.197 within 1.072 seconds in AWGN channel.
Since our main focus is on low SNR region, another possible The Rayleigh fading represents the effect of heavily built-
way of minimizing the total error rate is by increasing the up urban environments on radio signals. As in Fig. 5, neither
number of samples. In the spectrum sensing, IEEE 802.22 at -15 dB nor at -20 dB can the requirements be met within 2
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2011 proceedings

2 ∗
∗ e (λ )
seconds (or equivalently at N ≤ 125000), e.g., Pmd ≈ 0.293 Using (7), it is easy to show that ∂ P∂λ 2 > 0, and
when N = 125000 at -20 dB. It needs more than 3×106 sam- therefore, there is a global minimum at λ = λ∗ .
∗
ples to achieve Pmd ≤ 0.1 at -20 dB. Therefore, one possible
R EFERENCES
way of detecting the signal within 2 seconds is by increasing
the sampling frequency fs beyond 1.5 MHz. Another possible [1] C. Stevenson, G. Chouinard, Z. Lei, W. Hu, S. Shellhammer, and
W. Caldwell, “IEEE 802.22: The first cognitive radio wireless regional
way is to use a diversity technique or cooperative spectrum area network standard,” IEEE Commun. Mag., vol. 47, no. 1, pp. 130–
sensing topology to increase the number of effective decision 138, Jan. 2009.
statistics while maintaining a small number of samples per [2] H. Urkowitz, “Energy detection of unknown deterministic signals,” Proc.
of the IEEE, vol. 55, no. 4, pp. 523–531, Apr. 1967.
node or per branch. [3] V. Kostylev, “Energy detection of a signal with random amplitude,” in
VI. C ONCLUSION Proc. IEEE ICC’02, pp. 1606–1610.
[4] F. F. Digham, M. S. Alouini, and M. K. Simon, “On the energy detection
Detection performance of the energy detector is stud- of unknown signals over fading channels,” IEEE Trans. Commun.,
ied in the low SNR regime. The average missed-detection vol. 55, no. 1, pp. 21–24, Jan. 2007.
[5] S. Atapattu, C. Tellambura, and H. Jiang, “Energy detection of primary
probability is derived in a closed-form over Rayleigh and signals over η − μ fading channels,” in Proc. ICIIS, Dec. 2009, pp.
Nakagami-m fading channels. For minimizing the total error 118–122.
rate, the optimal detection threshold is exactly derived for [6] S. Atapattu, C. Tellambura, and H. Jiang, “Performance of an energy
detector over channels with both multipath fading and shadowing,” IEEE
AWGN channel, and it is also approximated for Rayleigh Trans. Wireless Commun., vol. 9, no. 12, pp. 3662–3670, Dec. 2010.
and Nakagami-m fading channels. The false alarm and the [7] H. Sun, D. I. Laurenson, and C.-X. Wang, “Computationally tractable
missed-detection probabilities may not be satisfied even at the model of energy detection performance over slow fading channels,”
IEEE Commun. Lett., vol. 14, no. 10, pp. 924–926, Oct. 2010.
optimal detection threshold. Increasing the number of samples [8] S. Atapattu, C. Tellambura, and H. Jiang, “Energy detection based
until the maximum allowable sensing time is achieved, and cooperative spectrum sensing in cognitive radio networks,” IEEE Trans.
increasing the sampling frequency are two possible ways to Wireless Commun., 2011, to appear.
[9] S. Atapattu, C. Tellambura, and H. Jiang, “Analysis of area under the
meet the requirements of the false-alarm and the missed- ROC curve of energy detection,” IEEE Trans. Wireless Commun., vol. 9,
detection probabilities. The research findings help to design no. 3, pp. 1216–1225, Mar. 2010.
an energy detector in an implementable way as fulfilling [10] S. Atapattu, C. Tellambura, and H. Jiang, “Performance of energy detec-
tion: A complementary AUC approach,” in Proc. IEEE GLOBECOM,
fundamental sensing requirements proposed in IEEE 802.22 Dec. 2010.
WRAN (channel detection time ≤ 2 seconds, missed-detection [11] Y.-C. Liang, Y. Zeng, E. Peh, and A. T. Hoang, “Sensing-throughput
and false alarm probabilities ≤ 0.1). tradeoff for cognitive radio networks,” IEEE Trans. Wireless Commun.,
vol. 7, no. 4, pp. 1326–1337, Apr. 2008.
A PPENDIX [12] R. Tandra and A. Sahai, “SNR walls for signal detection,” IEEE J. Select.
∂Pe (λ) Topics in Signal Processing, vol. 2, no. 1, pp. 4–17, Feb. 2008.
The optimal threshold, λ∗ , is achieved when ∂λ = [13] Z. Quan, S. Cui, A. Sayed, and H. Poor, “Optimal multiband joint
0. With the aid of (1), (2), Q(x) = 12 Erfc x
√
2
, and detection for spectrum sensing in cognitive radio networks,” IEEE Trans.
Signal Processing, vol. 57, no. 3, pp. 1128–1140, Mar. 2009.
(x−a)2
 x−a  −
b2
[14] C. Cordeiro, K. Challapali, D. Birru, and S. Shankar N, “IEEE 802.22:
∂
∂x Erfc b = − 2e √
b π
, we have An introduction to the first wireless standard based on cognitive radios,”
Journal of Commun. (JCM), vol. 1(1), pp. 38–47, Apr. 2006.
2 )2
(λ−N (1+γ)σw 2 )2
(λ−N σw
− − [15] S. J. Shellhammer, “Spectrum sensing in IEEE 802.22,” in First IAPR
∂Pe (λ) e 2N (1+2γ)σw4 e 4
2N σw
Workshop on Cognitive Information Processing, 2008.
= − √ = 0. (10) [16] M. K. Simon and M.-S. Alouini, Digital Communication over Fading
∂λ 2πN (1 + 2γ)σw2 2πN σw 2
Channels, 2nd Ed. New York: Wiley, 2005.
After some algebraic manipulations and taking the logarithm, [17] A. P. Prudnikov, Y. A. Brychkov, and O. I. Marichev, Integrals and
Series, vol. 2. Gordon and Breach science Publishers, 1986.
(10) can be simplified into a quadratic equation of λ as [18] W. Zhang, R. Mallik, and K. Letaief, “Optimization of cooperative
4
  spectrum sensing with energy detection in cognitive radio networks,”
N σw (1 + 2γ) ln(1 + 2γ)
λ2 − N σ w2
λ− Nγ + = 0. IEEE Trans. Wireless Commun., vol. 8, no. 12, pp. 5761–5766, Dec.
2 γ 2009.
[19] Z. Quan, S. Cui, A. Sayed, and H. Poor, “Wideband spectrum sensing
Thus, the solution for λ is in cognitive radio networks,” in Proc. IEEE Int. Conf. Commun. (ICC),
    2008, pp. 901–906.
2
N σw (1 + 2γ) ln(1 + 2γ) [20] Z. Quan, S. Cui, and A. Sayed, “Optimal linear cooperation for spectrum
λ= 1 ± 1 + 2γ 1 + .
2 N γ2 sensing in cognitive radio networks,” IEEE J. Select. Areas Signal
Processing, vol. 2, no. 1, pp. 28–40, Feb. 2008.
Since λ ≥ 0, λ∗ can be selected as (7). In low SNR, i.e., γ  [21] R. Fan and H. Jiang, “Optimal multi-channel cooperative sensing in
cognitive radio networks,” IEEE Trans. Wireless Commun., vol. 9, no. 3,
1, 1 + 2γ ≈ 1 and thus ln(1 + 2γ) ≈ 0, the optimal threshold pp. 1128–1138, Mar. 2010.
can be well approximated as (8). Now we can consider the [22] R. Fan, H. Jiang, Q. Guo, and Z. Zhang, “Joint optimal cooperative sens-
second order derivative which is given as ing and resource allocation in multichannel cognitive radio networks,”
IEEE Trans. Veh. Technol., 2011, to appear.
2 )2
(λ−N σw
− [23] Q. Peng, P. C. Cosman, and L. B. Milstein, “Optimal sensing disruption
∂ 2 Pe (λ) (λ − N σw
2
)e 4
2N σw
for a cognitive radio adversary,” IEEE Trans. Veh. Technol., vol. 59,
= √
∂λ2 2πN 3/2 σw
6 no. 4, pp. 1801–1810, May 2010.
[24] D. Cabric, A. Tkachenko, and R. W. Brodersen, “Experimental study of
2 )2
(λ−N (1+γ)σw
− spectrum sensing based on energy detection and network cooperation,”
2
(λ − N (1 + γ)σw )e 2N (1+2γ)σw4 in Proc. First Int. Workshop on Technology and Policy for Accessing
− √ . Spectrum (TAPAS), Aug. 2006.
2πN 3/2 σw
6 (1 + 2γ)3/2