See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/261263601

Adaptive threshold for energy detector based on Discrete Wavelet Packet

Transform

Conference Paper · April 2012

DOI: 10.1109/WTS.2012.6266117

CITATIONS READS

7 65

4 authors, including:

Zhijin Qin Nan Wang

Queen Mary, University of London China Telecom Corporation Limited
55 PUBLICATIONS 651 CITATIONS 6 PUBLICATIONS 78 CITATIONS

SEE PROFILE SEE PROFILE

Yue Gao
Queen Mary, University of London
168 PUBLICATIONS 1,596 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

IoT for smart cities View project

Physical layer security View project

All content following this page was uploaded by Zhijin Qin on 07 November 2015.

The user has requested enhancement of the downloaded file.

 Adaptive Threshold for Energy Detector Based on
Discrete Wavelet Packet Transform

Zhijin Qin Nan Wang, Yue Gao and Laurie Cuthbert

Beijing University of Posts and Telecommunications Electronic Engineering and Computer Science
Queen Mary University of London Queen Mary University of London
Beijing, China London, UK
qinzhijin@gmail.com {nanw}{ yue.gao}{laurie.cuthbert}@eecs.qmul.ac.uk

Abstract—One promising approach to achieving high precision their later work, a double threshold is used, and the channels
and low complexity in a cognitive radio system is the energy can be classified into three states: the assured occupied channel
detector based on the Discrete Wavelet Packet Transform (black), the unassured channel (gray), and the assured
(DWPT). However, the thresholds used in previous spectrum unoccupied channel (white) [6]. The energy detector will only
detection algorithms are fixed values determined by false alarm transfer the unassured channels to the feature detector at the
probability and sample points. When the Signal to Noise Ratio second stage. As a result, the number of channels to be
(SNR) is lower than 5dB, an energy detector with fixed threshold processed at the second stage was reduced, and the
may falsely determine some unoccupied sub-channels as computational complexity was decreased as well.
occupied. In this paper, a threshold adapter adopting adaptive
threshold is proposed to increase the detection sensitivity, The performance of the energy detector is mainly
meanwhile, reduce the computational complexity. Simulation dependent on channel environments such as noise, interference,
results are presented to demonstrate the effectiveness of this shadowing and multipath fading [7]. Thus, setting a proper
algorithm. threshold for the energy detector is the key to obtaining higher
sensitivity. Spectrum sensing with a fixed threshold works well
Keywords- spectrum sensing, discrete wavelet packet transform, when the SNR is high; however, when channel environment
fixed threshold, adaptive threshold becomes poor, for instance, the SNR becomes lower, the
energy of each channel increases as a function of noise
I. INTRODUCTION increasing. If the threshold stays the same as the value when
With advanced developments in wireless communication the SNR is high, the energy of each channel will become much
services, efficient use of spectrum is becoming more higher than it. As a result, some unoccupied sub-channels will
significant. Cognitive radio was proposed to address this issue be determined as occupied, which makes secondary users lose
by fully making use of radio resources [1]. It can sense the chances to access these white channels.
environment and serve target users without causing
interference to primary users [2]. Spectrum sensing is the first II. Energy Detector Based on Discrete Wavelet Packet
step in detecting the surrounding spectrum environment in a Transform
cognitive radio system. There are several kinds of techniques
The concept of DWPT and the model of energy detector are
used for spectrum detection, such as matched filter detection,
introduced in this section.
energy detection, and cyclostationary feature detection. Energy
detection is the most common scheme of spectrum sensing
with low computational complexity [3]. Cyclostationary A. Discrete Wavelet Packet Transform
feature detection is used to detect the primary users by With DWPT, a signal can be represented as
calculating the cyclostationary features of target signals. It is
more precise than energy detector, but requires higher f (t ) = ∑ ∑ (c
j ≥ j0 k
j,k ϕ j , k (t ) + d j , kψ j , k (t )) (1)
computational complexity.
To obtain high precision with low computational where c j , k and d j , k are scaling and wavelet coefficients
complexity, a two-stage sensing architecture has been proposed respectively. ϕ j , k (t ) and ψ j , k (t ) are wavelet bases. ϕ j , k (t )
by the IEEE 802.22 working group [4]. At the first stage, the
is scaling function describing approximation of a signal, and
energy detector is used to determine whether the spectrum is
occupied or not. At the second stage, the feature detector is ψ j , k (t ) is wavelet function describing detailed space of a
used to determine whether these unoccupied spectrum are signal. Energy of signals can be measured as follows: [8]
occupied by weak signals. Youngwoo Youn et al [5] used an
1 T 2
energy detector to obtain the energy of each sub-channel, and E= ∫ [ ∑ ∑ c j , k ϕ j , k (t ) + d j , kψ j , k (t )] dt (2)
then it was compared with a fixed threshold to determine T 0 j ≥ j0 k
whether the channel is occupied by primary users or not. In

978-1-4577-0580-9/12/$26.00 c2012 IEEE

 1 unoccupied. Then the channel will not be processed in
=
T
∑ ∑ (c
j ≥ j0 k
j,k
2
+ d j,k 2 ) the next level DWPT decomposition. If the energy is
higher than the threshold, it will be processed in the next
level DWPT decomposition. Here, the threshold is
B. Energy Detector determined by the threshold adapter. In the first level
The procedure for the energy detector based on DWPT is DWPT decomposition, fixed threshold is applied and
described as follows: calculated by (3). The adaptive threshold will be used
1) perform n levels DWPT to the received signals; for the further level DWPT decomposition.
z Steps 2 and 3 are repeated until the necessary iteration
2) calculate the energy of each sub-channel; (RI) levels.
3) compare it with the fixed threshold and classify
sub-channels as occupied or unoccupied. When doing (i+1)th-level DWPT decomposition to the
spectrum of interest, the adaptive threshold ( λi +1, j ) for
The fixed threshold ( λ ) is determined by the false alarm
channel j can be calculated by
probability ( Pf ) and the number of sample points (N). It can
be calculated as follows: λi , j + α ⋅ Ei +1,2 j
λi +1,2 j = (4)
λ = Q −1 ( Pf ) 2 N + N (3) 2
λi , j + α ⋅ Ei +1,2 j −1
1 ∞ u2 λi +1,2 j −1 = (5)
where Q( x) =
2π
∫x
exp(− )du .
2
2
where i, j are index. Ei +1,2 j is the energy of channel 2 j in the
III. PROPOSED ALGORITHM spectrum of interest after (i+1)th-level DWPT decomposition.
Channel j is one of the sub-channels when doing ith-level
In order to classify spectrum of interest more precisely DWPT decomposition to the spectrum of interest. Channels
and reduce computational complexity, a new algorithm is 2 j − 1 and 2 j are the further divisions of channel j after
proposed in this section. In the new algorithm, the threshold
adapter is used to obtain an adaptive threshold during the (i+1)th-level DWPT decomposition. λi , j , λi ,2 j −1 and λi ,2 j are
process of classifying the spectrum of interest. thresholds for channel j , channel 2 j − 1 and channel 2 j
The procedure of the proposed algorithm is shown in respectively. α (0<α<1) is the parameter to determine the
Figure 1 and described as follows: adaptive threshold.
In (4) and (5), when α tends to 1, the adaptive threshold is
affected largely by the energy of the received signals. When α
tends to 0, the threshold for each sub-channel after the (i+1)th-
level DWPT decomposition tends to be half of the threshold
after the ith-level DWPT decomposition. It is also similar to the
fixed threshold of sub-channels after the (i+1)th-level DWPT
decomposition.

Ei , j λi , j IV. SIMULATION ENVIRONMENT AND RESULTS

In the simulation process, we assumed that 16 sub-
channels are present in the spectrum of interest (64 MHz) and
Ei, j λi , j the bandwidth for each sub-channel is 4 MHz. So the 4-level
DWPT decomposition is applied. Since the down-sampling is
required at each level of DWPT decomposition, the order of
wavelet packets in higher frequency and lower frequency is
exchanged at every level DWPT decomposition. Therefore,
these wavelet packets will not be sorted from low frequency to
Figure 1. Procedure of Proposed Algorithm high frequency. The frequency range of each wavelet packet is
shown in following table.
z Compute the necessary iteration RI for DWPT
as RI = log 2 ( Bs / Bd ) . Bs refers to the spectrum of
TABLE I. FREQUENCY RANGE OF EACH WAVELET PACKET
interest and Bd to the bandwidth of each sub-channel.
Packet Frequency Packet Frequency
z Perform ith-level DWPT decomposition (the initial value number range (MHz) number range (MHz)
of i equals to 1).
z Calculate the energy of each channel, and compare it (4, 0) (0, 4) (4,1) (4, 8)
with the threshold. If the energy is lower than the (4,2) (12, 16) (4,3) (8, 12)
threshold, the related channels are determined as
 (4,4) (28, 32) (4,5) (24,28) threshold of each sub-channel has been further enlarged and
(4,6) (16, 20) (4,7) (20, 24) sub-channels without primary users can be detected as
unoccupied. In Figure 5, taking α=0.55, the adaptive threshold
(4,8) (60, 64) (4, 9) (56, 60)
become higher than the energy of sub-channels 3 and 7. As a
(4,10) (48, 52) (4,11) (52, 56) result, these two channels will be determined as unoccupied
(4,12) (32, 36) (4,13) (36, 40) even though primary users exist in them. As a result, secondary
users can access to these channels, which would lead to
(4,14) (44, 48) (4,15) (40, 44) harmful interference to the primary users.

The false alarm probability is set to be 0.01. The primary

5
x 10
3
users are assumed as 3 MHz bandpass signals, and there are
three channels are occupied by primary users. In the simulation 2.5
process, the centre frequencies of the three primary users are
fixed as f = 9, 27, 45 MHz, respectively. Therefore, the energy 2
of sub-channel 3 (8 MHz, 12 MHz), 7 (24 MHz, 28 MHz), 12

Power (in Decimal)

Simulation Energy
Adaptive Threshold
(44 MHz, 48 MHz) is anticipated to be higher than other 1.5
Fixed Threshold

channels after 4 levels DWPT decomposition.

As assumed, there is no primary user in sub-channel 2. 1

However, when SNR=5dB, sub-channel 2 will be falsely

determined as occupied by the fixed threshold. This is shown 0.5

in Figure 2. This will lead the secondary users losing the

chance to access the spectrum from 4 MHz to 8MHz. When 0
0 2 4 6 8 10 12 14 16
SNR is lower than 5 dB, the false alarm probability will Sub Channels after 4th Level DWPT Decomposition

increase. While when SNR is higher than 5dB, the fixed

threshold can be used to accurately determine whether the Figure 3. Energy of Sub-channels, α=0.45, SNR=-17dB
channel is occupied or not. 5
x 10
4
3
x 10
6
Simulation Energy
Fixed Threshold 2.5
5

2
Power (in Decimal)

Simulation Energy
4 Adaptive Threshold
Power (in Decimal)

Fixed Threshold
1.5
3

1
2

0.5
1

0
0 2 4 6 8 10 12 14 16
0 Sub Channels after 4th Level DWPT Decomposition
0 2 4 6 8 10 12 14 16
Sub Channels after 4th Level DWPT Decomposition
Figure 4. Energy of Sub-channels, α=0.50, SNR=-17dB
Figure 2. Energy of Sub-channels, SNR=5dB
5
x 10
3
Then we adopt the adaptive threshold to detect the primary
users in the spectrum of interest. Set the SNR of received 2.5
signals as -17dB, and take α=0.45, 0.50, 0.55. Before
calculating the energy of each sub-channel, the wavelet packets 2
Power (in Decimal)

are re-ordered from low frequency to high frequency. After Simulation Energy
Adaptive Threshold
performing the 4-th level DWPT decomposition, the energy of 1.5 Fixed Threshold

each sub-channel is shown in Figure 3 to Figure 5.

In Figure 3, it shows the fixed threshold is much lower 1

than the simulation energy of each sub-channel. However,

when the threshold adapter is adopted, the new threshold will 0.5

be enlarged. But the thresholds in all sub-channels are still

lower than the simulation energy. Therefore, the case of α=0.45 0
0 2 4 6 8 10 12 14 16
cannot improve the performance. In Figure 4, when α=0.50, the Sub Channels after 4th Level DWPT Decomposition

detection sensitivity has been improved because the adaptive

Figure 5. Energy of Sub-channels, α=0.55, SNR=-17dB
 4 Simulation Energy
x 10
8 Adaptive Threshold
When α=0.53, the energy of sub-channel 3, 7, 12 is higher Fixed Threshold
than the adaptive threshold, as shown in Figure 6. If we keep α 7

increasing, the energy of the three sub-channels will be lower 6

than the adaptive threshold. As a result, these channels will be
falsely determined as unoccupied. Therefore, the proper range

Power (in Decimal)

5

for α is from 0.5 to 0.53.

4
5
x 10
3 3

2
2.5

2
Power (in Decimal)

Simulation Energy 0
Adaptive Threshold 1 1.5 2 2.5 3 3.5 4
Fixed Threshold Sub Channels after 2 Level DWPT Decomposition
1.5

(b)
1
4
x 10 Simulation Energy
7 Adaptive Threshold
0.5 Fixed Threshold
6

0
0 2 4 6 8 10 12 14 16 5
Sub Channels after 4th Level DWPT Decomposition

Power (in decimal)

4
Figure 6. Energy of Sub-channels, α=0.53, SNR=-17dB
3
Finding out the unoccupied channel as early as possible
can reduce the computational complexity as the occupied 2

channel will not be processed in the further level DWPT

decomposition. Thus, the threshold should be as large as 1

possible. So the optimal value of α is 0.53.

0
1 2 3 4 5 6 7 8
In the case of α=0.53 and SNR= 0dB, the 4th-level DWPT Sub Channels after 3 Level DWPT Decomposition

decomposition of the received signals is shown in Figure 7.

After the 2nd-level DWPT decomposition, the energy of sub- (c)
channel 4 (48MHz, 64MHz) is higher than the fixed threshold. x 10
4

So, if the fixed threshold is used, sub-channel 4 is determined 6

as occupied and should be processed at the 3rd-level DWPT

decomposition. If the adaptive threshold is applied, the energy 5 Simulation Energy

of sub-channel 4 becomes lower than the threshold. So it will Adaptive Threshold

Fixed Threshold
be determined as unoccupied and will not be processed at the 4
Power (in decimal)

next level DWPT decomposition. Hence, the computational

complexity of the sensing algorithm is reduced and secondary 3

users can access the spectrum between 48 MHz and 64 MHz.

2
4
x 10
15
Simulation Energy
1
Adaptive Threshold
Fixed Threshold

0
0 2 4 6 8 10 12 14 16
10 Sub Channels after 4 Level DWPT Decomposition
Power (in Decimal)

(d)
Figure 7. Energy of Sub-channels, α=0.53, SNR=0dB (a) 1st Level DWPT
5
Decomposition (b) 2nd Level DWPT Decomposition (c) 3rd Level DWPT
Decomposition (d) 4th Level DWPT Decomposition

V. CONCLUSTIONS
0
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8
Sub Channels after 1 Level DWPT Decomposition
1.9 2 In this paper, we proposed a threshold adapter to get an
adaptive threshold for an energy detector based on DWPT. The
(a) threshold of a channel changes adaptively as a function of the
energy of that channel. It is an efficient perspective to
classifying spectrum as it can improve performance of the
energy detector when the SNR is low. Based on the simulation
results, the fixed threshold cannot detect primary users
accurately when the SNR is lower than 5 dB. However, the
threshold adapter can accurately detect out the primary users
when the SNR is -17 dB. Therefore, a 22 dB gain of detection
sensitivity has been obtained. In addition, once a spectrum has
been detected as unoccupied at a lower level DWPT
decomposition, it does not need to be processed in the next.
This will reduce the computational complexity.

REFERENCES
[1] S. Haykin, “Fundemental issues in cognitive radio,” in Cognitive
Wireless Communication Networks, E. Gissaub abd V. K. Bhargava, Eds,
New York: Springer, 2007.
[2] I.J. Mitola and G. Q. Maguire, “Cognitive radio: making softwave radios
more personal,” in IEEE Personla Commun, vol. 6, pp. 13-18, Aug.
1999.
[3] S. Jeng, J. Chen, H. Lin and C. Tsung, “Wavelet-Based Spectrum
Sensing for Cognitive Radios using Hilbert Transform”, in World
Academy of Science, Engineering and Technology , 2011.
[4] IEEE 802.22 WG, “Functional Requirements for the IEEE 802.22
WRAN standard,” Sep. 2005.
[5] Y. Youn, Y. Jeon, H. Jung and H. Lee, “Discrete Wavelet Packet
Transform based Energy Detector for Cognitive Radios”, in IEEE
Vehicular Technology Conf. Spring, 2641-2645, Apr. 2007.
[6] S. Kim, Y. Yoon, H. Jeon, M. Kim, and H. Lee, “Selective discrete
wavelet packet trnasform-based energy detector for cognitive radios”, in
MILCOM 2008. IEEE, San Diego, 16-19 Nov. 2008.
[7] J. Lee, J. Kim, H. Oh, and S.Hwang, “Energy Detector using Adaptive –
Fixed Thresholds in Cognitive Radio Systems ”, in Aisa Pacific
Communication Conference 2008.
[8] W. Yoon and M. J. Devaney, “Power Measurement Using the Wavelet
Transform”, in IEEE Transactions on Instrumentation and Measurement,
vol.47, no. 5, Oct. 1998.

View publication stats