Professional Documents
Culture Documents
An Adaptive Threshold Based Energy Detector For Spectrum Sensing in Cognitive Radios at Low SNR
An Adaptive Threshold Based Energy Detector For Spectrum Sensing in Cognitive Radios at Low SNR
.
The test statistic is given by
) ( y Z =
N
1
*
=
N
n 1
|y[n]|
2
(3)
For large N, the test statistic Z(y) has a normal distribution
with mean
i
and variance
i
under hypothesis H
i
(i = 0,1) [5].
The mean and variance of the test statistic have been shown,
using the results in [9, 10]:
0 =
2
n (4)
0 =
N
n
2
(5)
1 =
2
n *(1+ ) (6)
1 = 0 *
) 1 2 ( +
(7)
The probability of false alarm, P
fa,
of such a system is given by
) 0 | ) ( ( H y Z P Pfa > = (8)
For a sufficiently large number of samples, N, P
fa,
is [9]:
fa P =Q(
0
0
) (9)
Similarly the probability of detection, P
d,
of such a system is
given by
BPF ADC ( )
2
T
Z(y)>
or
Z(y)<
575
) 1 | ) ( ( H y Z P Pd > = (10)
P
d
can be expressed as
d P =Q(
1
1
) (11)
where Q(x) is defined by the complementary distribution
function of the standard Gaussian and is given as
Q(x) =
x
u
du e
2 /
2
2
1
(12)
As seen from (9) and (11), the threshold, , can be set for a
target P
fa
or P
d.
For the sake of notational simplicity we will
denote a threshold derived using (9) as
f
and the one derived
using (11) as
d
. It is clear from (6), (7) and (11) that setting a
threshold based on target P
d
requires information about , i.e.
SNR of the channel.
B. The Problem Statement
An energy detector based on threshold calculated for a
target P
fa
works well in high conditions. But when is less
than -15dB, the performance of the energy detector leaves
much to be desired [4]. On the other hand, if we assume that
of the channel is available, then it can be used to calculate the
threshold for a target probability of detection using (11).
(a)
(b)
Fig. 2.(a) Probability of detection (b) Probability of false alarm of energy
detector for the two thresholds f and d at a SNR of -15dB. Target Pd = 0.9
and Pfa =0.01
The performance of such a detector is steady and meets the
target probability of detection in low scenarios but the
tradeoff is that the corresponding P
fa
also increases beyond the
targeted value. The simulation results to demonstrate this are
shown in Fig. 2. Since the knowledge of is not available, at
least initially when the system powers ON, the conventional
energy detectors calculate their threshold based on target P
fa
[3] by using (9) and then observe the probability of detection
they can manage for the system. This, as explained above, has
the disadvantage that if is less than the expected range, the
performance is severely affected.
From Fig. 2 we can see that as long as the sensing time is
short and consequently the number of samples less than N
c
, a
conventional
f
threshold based detector will give a very poor
probability of detection metric. Thus, in practical scenarios
where increasing the sensing time to get the desired number of
samples, N
c
, is not feasible, we may have to choose some
other option.
In the next section, we explain how a strategy can be
devised to change the threshold from one based on P
fa
to a
threshold based on P
d
to better the performance metric.
III. PROPOSED SYSTEM MODEL
A. System Block Diagram and Analysis
Fig. 3. Block diagram of proposed energy detector using adaptive setting of
threshold.
Fig. 3 shows the block diagram of the proposed energy
detector. One of the key units is the SNR estimator which is
used to give an estimate of the channel condition. This
information coupled with other parameters, as explained later,
would allow the Threshold Setter module to choose the right
threshold so as to optimize the performance metric.
Using (9) and (11) and substituting the mean and variance
values from (4) to (7), it can be easily shown that sensing
time, or in other words, the number of samples, N, is related to
P
fa
, P
d
and , as
[ ]
2
1 1
2
1 2 * ) ( ) (
1
+ =
d fa c P Q P Q N (13)
This derived equation matches with the one in [9]. The
significance of this is that it gives the critical number of
BPF ADC ( )
2
T
Threshold
Setter
SNR
Estimator
576
samples, N
c
, required to achieve the target metrics (P
d
& P
fa
).
In other words, by varying the sensing time of the signal and
choosing an appropriate threshold depending on the sensing
time, we can achieve a result to suit the requirements.
Knowing the critical samples to achieve the target P
d
& P
fa
,
we can use it to decide on how best to switch the threshold
from one value to another. To do this, we introduce a control
parameter, , which can be changed to vary the threshold from
f
to
d
. The threshold
is given as
) ( f d f + = , 1 0 (14)
At a simple level, can take a binary value of 0 or 1, thereby
changing the threshold from
f
to
d
. The results for other
values of have also been simulated. Please note that the
value of chosen here are for illustration purposes only. Any
value of within the range [0,1] can be chosen.
B. The Adaptive Approach
Having established the system model, we now present the
approach that can be taken to improve upon the performance
metrics in low channel conditions. We suggest that the
cognitive radio can set the threshold to
f
on startup and then
immediately do an estimation of the channel . Then using
(13) it can calculate the critical samples it needs to reach the
target performance metrics. If the operational requirement
allows the cognitive radio to sense the channel long enough to
get the required number of samples, i.e. N N
c
then the
threshold can be maintained at
f
. Or if the requirement is
more stringent on maximizing the throughput of the system,
then the threshold can be switched over to
d
by changing to
1. On the other hand, if the practical requirements do not allow
the cognitive radio to sense for long, i.e. N < N
c,
then by
changing suitably, the desired response can be obtained
albeit at the cost of the other performance metric.
IV. DISCUSSION
An analysis of the response given in Fig. 4 brings the
following into light.
1. As the value of reaches changes from 0 to 1, the curve
obtained moves away from the graph obtained for
f
to the one
obtained for
d
.
2. There is always a tradeoff involved in choosing one
metric over the other.
3. The best response of the energy detector is depended on
the channel conditions, the sensing time available on hand and
the operational requirements.
Let us elaborate on the third aspect to see the merits of
having an adjustable parameter to suit the operating
environment. As discussed earlier, a cognitive radio system
can have two broad but conflicting operational requirements.
It could be either to cause least interference to the primary
user (thus necessitating a high P
d
) or to maximize the
throughput of the cognitive radio to utilize the channel to the
maximum extent (thereby requiring a very low P
fa
) but at the
risk of interfering with the primary user. The algorithm given
in Fig. 5 is for an environment where detecting the PU is the
(a)
(b)
Fig.4.(a) Probability of detection (b) Probability of false alarm of energy
detector using adaptive setting of threshold for various values of at SNR of -
15dB and target Pd and Pfa of 0.9 and 0.01 respectively.
more important metric than the throughput of the cognitive
radio system. In such a situation, keeping the probability of
detection high should be of more concern than the false alarm
probability.
Fig.5. Algorithm to keep the probability of detection of primary user at high
levels by varying .
Estimate SNR
Is N
> N
c
Calculate N
c
> 0, =
f
> 1, =
d
Input Samples
N
Yes
No
577
Thus, if the system finds that it is has to operate with less
number of samples than the critical samples, N
c
, due to time
constraints, then the cognitive radio system can opt for
d
as
the threshold to differentiate the test statistic and thus manage
to keep the detection chances high. The algorithm is suitable
and can be easily modified for a cognitive radio system
requiring high throughput. If the requirement is to have an
average detection probability and average false alarm
probability, then can be set to 0.5 and thus a compromise
can be achieved which would otherwise be not possible in a
conventional fixed threshold based energy detector.
V. CONCLUSION
Setting the threshold level of an energy detector is the key
parameter that determines its performance metrics vis a vis the
ability to detect the primary user correctly and the ability to
discriminate a false alarm. Ideally an energy detector should
have a high probability of detection and a very low probability
of false alarm, but practically there is always a tradeoff
between the two parameters.
In this paper we have shown how the tradeoff can be
managed in the best possible way by using a adaptive
threshold. We present a new perspective to the role of the
sensing time in determining the performance metric in low
SNR channels. We have also shown that the critical number of
samples required should be calculated by the system by
estimating the SNR so as to determine how best to vary the
threshold to get the desired curve for the given operating
environment. Further research can be done to compare the
performance of the CR system with other adaptive techniques
to show the merits of the proposed technique.
REFERENCES
[1] Federal Communications Commission, "Notice of proposed rule making
and order: Facilitating opportunities for flexible, efficient, and reliable
spectrum use employing cognitive radio technologies", ET Docket No
03-108, Feb. 2005-73.
[2] D. Cabric, S. Mishra, and R. Brodersen, "Implementation issues in
spectrum sensing for cognitive radios" in Proc. Asilomar Conf. on
Signals, Systems and Computers, Nov. 2004, vol. 1, Pacific Grove,
California, USA, pp. 772776.
[3] T Yucek and H Arslan, "A Survey of Spectrum Sensing Algorithms for
Cognitive Radio Applications", IEEE Communications Surveys &
Tutorials, vol. 11, No. 1, pp. 116-130, 2009.
[4] A. Sahai, N. Hoven, and R. Tandra, "Some fundamental limits on
cognitive radio", in Proc. Allerton Conf. on Commun., Control, and
Computing, Oct. 2004, Monticello, Illinois.
[5] H. Urkowitz, "Energy detection of unknown deterministic signals",
Proc. IEEE, vol. 55, pp. 523531, Apr. 1967.
[6] Zhuan Ye, G Memik, John Grosspietsch, "Energy Detection using
Estimated Noise Variance for Spectrum Sensing in Cognitive Radio
Networks", in Proc. Wireless Communication and Networking
Conference, April 2008, Las Vegas, NV, pp. 711 - 716..
[7] A Gorcin, Khalid A Q,H Celebi, H Arslan, "An Adaptive Threshold
Method for Spectrum Sensing in Multi Channel Cognitive Radio
Networks", in Proc. IEEE 17th International Conference on
Telecommunications (ICT), April 2010, Doha, Qatar, pp. 425-429.
[8] H H Choi, K Jang, Y Cheong, "Adaptive Sensing Threshold Control
Based on Transmission Power in Cognitive Radio Systems",
Cognitive Radio Oriented Wireless Networks and Communications, May
2008, Singapore, pp. 1 -6.
[9] Y C Liang, Y Zeng, E C Y Peh, A T Hoang, "Sensing Throughput
TradeOff for Cognitive Radio Networks", IEEE Transactions on
Wireless Communications, vol 7, No.4, April 2008, pp. 1326-1337.
[10] A.Sonnenschien and P.M.Fishman, "Radiometric Detection of Spread
Spectrum Signals in Noise of Uncertain Power", IEEE Transactions on
Aerospace and Electronic Systems, volume: 28 , No: 3, July 1992, pp.
654 - 660.
[11] J W Lee, J H Kim, H J Oh and S H Hwang, "Energy Detector using
Adaptive-Fixed Threshold in Cognitive Radio Systems", In Proc. 14th
Asia-Pacific Conference on Communications, Oct. 2008, Tokyo, Japan,
pp. 1-4.
[12] Shi et al, United States Patent Pub. No: US 2010/0081387 A1, Apr. 1,
2010.
578