 Open Access
 Total Downloads : 382
 Authors : Arun. A. Badiger, H. C. Srinivasaiah
 Paper ID : IJERTV2IS70245
 Volume & Issue : Volume 02, Issue 07 (July 2013)
 Published (First Online): 09072013
 ISSN (Online) : 22780181
 Publisher Name : IJERT
 License: This work is licensed under a Creative Commons Attribution 4.0 International License
Techniques For Detection Of Analog TV Whitespace In Electromagnetic Spectrum For Cognitive Radio Applications
Techniques For Detection Of Analog TV Whitespace In Electromagnetic Spectrum For Cognitive Radio Applications
Arun. A. Badiger #1, H. C. Srinivasaiah *2,
# Department of Telecommunication Engineering, Dayananda Sagar College of Engineering, Bangalore.
* Professor, Department of Telecommunication Engineering, Dayananda Sagar College of Engineering, Bangalore.
Abstract Spectrum sensing is the key to the success of cognitive radios (CRs), which detect whether a primary user is utilizing his allocated transmission channel or not, before itself launching transmission. Thus secondary users, who use CR, will use the primary users channel without harmfully interfering with his/her signal. In this paper, the most commonly used detection techniques namely Energy detection, Matched filter detection and Cyclostationary detection techniques are studied in the context of using analog TV channels for CR application. The cyclo stationary detection technique outperforms the other two by detecting the presence of primary users signal on a channel with signal to noise ratio (SNR) worse by more than 20dB, relative to the SNR thresholds of other two techniques. The evaluation of 3 detection techniques is done under a wide range of signal threshold () relative to noise. This study helps in gaining a deeper insight into the use of analog TV channels for CR applications.

INTRODUCTION
Due to switch over from analog to digital TV transmissions a wide range of whitespace is created in spectrum from 54 to 890 MHz. This white space can be dynamically allocated to users, to increase the throughput of transmission medium. When primary users have right over the allocated channel, secondary users can use the same without harmfully interfering with the primary transmission [1, 2]. Primary users are given the highest priority, and secondary users are given the next level priority. Todays wireless mobile communications have advanced significantly. Various kind of mobile services have increased the demand for high data rates for their users. The 3rd generation partnership project (3GPP) is working towards improved techniques to reach beyond the requirements; CR is one among those techniques.
The CR networks should be aware of their network, available spectrum environment and their location and capable to adapt to their environment by changing their transmission and reception parameters to support more
concurrent wireless communications in a spectrum band at one location.
To achieve confirming the vacant bands become the main issue, i.e. it is a must to ensure that, (1) channels which are classified for secondary use are actually not being utilized by primary users; and (2) the channels being used by secondary users should be freed and re allocated to primary user whenever they appear. The spectrum sensing is the core issue, which provide information regarding channel status.
The commonly used spectrum sensing techniques are energy detection technique, matched filtering technique and cyclostationary technique. In this paper we analyzed the performance of these three techniques, by comparing them in terms of probability of false alarm, probability of detection and probability of miss detection. The raising of false alarm indicates conclusion in favour of presence of primary users signal when its really not there. The raising of miss detection alarm indicates the conclusion in favour of absence of primary users signal when its really there. The conclusion of right detection indicates the true status of the channel. All these conditions are judged by testing null hypothesis (H0) and alternative hypothesis (H1), discussed later.
The remainder of the paper is organized as follows. In section 2 simulation of analog TV signal is introduced. In section 3, spectrum sensing methodologies are discussed. Section 4 discusses the results and comparisons of methods. In section 5 we conclude.

SIMULATION OF ANALOG TV SIGNAL
The Analog TV broadcast channels range from channel number 2 to 83, with a bandwidth ranging from 54 to 890MHz. The composite analog TV broadcast consists of video, audio and color signals. The color signal have low power spectral density compared to audio and video signals, so we have not considered the color signals for detection. A single composite Analog TV has a bandwidth of 6MHz.
The below Fig. 1 explains the generation of composite video signal, ma(t) is the audio signal mc(t) is the video signal. Audio signal is frequency modulated and video signal is first double side band suppressed carrier (DSBSC) modulated, later it is passed through filter to get vestigial side band (VSB) signal STV(t). Composite TV signal STV(t) is given as [3]:
() = 1 + 2 + () (1) where STV (t) is the composite video signal, (t) is frequency modulated audio signal centered at audio carrier, is the video signal amplitude. Frequency modulated audio signal is given by the equation
Fig. 2: Simulated block diagram of cognitive radio receiver.
Fig. 3 shows power spectral density at receiving antenna for the three TV channels namely channel number 2, 3 and 4 each with channel bandwidths of 5460 MHz, 60
66 MHz and 6672MHz. Fig. 4 shows the amplified signal of Fig. 3, with an overall gain of more than 60dB.
= Ã— cos 2 +
(2)
FM
Modulation
FM
Modulation
where is the amplitude of the audio signal and is frequency deviation constant, is composite audio signal.
ma (t)
Sa (t)
mc (t)
STV (t)
Fig.
DSBSC
Modulation
DSBSC
Modulation
VSB filter
VSB filter
1: Simulated system diagram of a broadcast TV transmitter.
is 114 KHz, and modulation index is 0.875 for visual TV signals. The signal is passed through Additive White Gaussian Noise (AWGN) channel and is received by wide band receiver. The Fig. 2 shows the Radio Frequency (RF) front end block diagram of CR receiver. The signal is captured by a dipole antenna, and is fed to the low noise amplifier (LNA), which will amplify the signal by suppressing the noise. Further signal is filtered by 2080 MHz wide bandwidth, using 5th order butter worth band pass filter with approximately 1dB loss [3]. Further this signal is amplified with nonlinear amplifier with gain of 25dB. Long coaxial cable introduces a loss of 13dB and a variable attenuator is used to restrict high voltage fluctuations. Further two stages of amplifiers with identical band pass characteristics are used with a band pass filter, in between them.
The two stage filters are used to reduce the nonlinearity effects. At last 12 bit analog to digital convertor (ADC) is used for digitization of the signal. The digitized signal is passed through sub channel filter to filter out only the required TV channel signal; this filter is a digital filter of 6MHz pass band. Now detection techniques can be
applied on this signal to obtain the channels status.
Fig. 3: Power spectral density of signal received at antenna.
Fig. 4: Power spectral density of the signal in Fig. 3 after amplification.

SPECTRUM SENSING METHODOLOGIES
The goal of the spectrum sensing techniques is to detect whether a particular TV channel spectrum is idle or not. Based on sensing, we decide whether a particular channel is used by its primary user or not. Before starting of usig a TV channel, we will not have any information regarding its status, whether idle or busy. We can make the following hypothesis:
=
+ 1
0
0
(3)
x(n) is the received signal, s(n) is the transmitted signal by the primary user i.e. TV broadcast signal, is path loss factor, w(n) is the zero mean AWGN of the channel. From Equation 3, 0 is represented as the null hypothesis where only noise is received by the receiver;
1 is the alternative hypothesis stating the presence of
of signal from noise will be very difficult. The test statistics for the energy detection are given below.
The probability of false alarm of the received signal
will be only noise , i.e. null hypothesis H0 is true [4]. Energy of the this signal can be estimated as,
=0
=0
TV signal, added with the AWGN and the same is
= 1  2
(7)
detected by at the CR receiver.
The detection methods results into following cases:
Probability of false alarm ( ): wherein 0 is true but
The energy detected will be chisquare (2) distribution with N degrees of freedom. It is given by below equation,
decision is made in favor of 1 , given by:
0 =
1
2
1
2 2 , 0 (8)
2
2
= ( / ) (4)
2 2 2 ()
1 0 2
Probability of detection ( ): wherein 1 is true and decision is made in favor of 1 given by:
= (1 / 1 ) (5)
Probability of miss detection ( ): wherein 1 is true and the decision is made in favor of 0 given by:
Noise x here is AWGN with mean = 0 and its variance
2 so that the probability of false alarm can be calculated as
= (1 /0 )
= 0 > /0
= (0 /1 ) (6)
The workings of the 3 sensing methods namely energy
= 0
detection, matched filter detection, and cyclo stationary detection technique are described in the following sections.
Solving the above equations we obtain the equation for probability of false alarm for energy detection as
, 2

ENERGY DETECTION METHOD
= 2 2
2
(9)
It is the simplest method among all the 3 detection techniques. Prior information regarding the primary user
where, is the threshold value, is the number of symbols used, 2 is variance of noise, () is the Gamma function which is give by
is not required for this method. This method just
= (1)
(10)
>
>
calculates the energy of the received signal and it is compared with a threshold value to make the decision. Fig. 5 explains the method.
0
and , is the upper incomplete gamma function,
which is given by the equation
Sub channel filter
Sub channel filter
()
1
()
, = (1)
(11)
ADC  2
Now for the probability of energy detection is given
=0
by = / . The alternative hypothesis is
1 1 1
Fig. 5: Block diagram indicating the procedure of Energy Detection
In this figure the digitized signal from the ADC output is fed to sub channel filtering which only filters the required TV channel signal. Now from output of this filter, we obtain the energy of the signal; this total energy is compared with the threshold. If the total energy is greater than the threshold then TV channel is being used by the primary user or else only noise is present and it can be allocated to the secondary user. This method does not require any prior knowledge of the primary signal, so it is easy to implement, and its computational complexity is less. But the noise power will be varying over each interval of time, it is very difficult to set always a same threshold value which makes its
performance low, and for very low SNR, differentiation
given in equation 3. The energy of the received signal is calculated as [4],
=0
=0
= 1 ()2, (12)
=0
=0
= 1  () + 2, (13)
where () is the transmitted signal, is path loss factor, and is AWGN. Hence the probability of detection for energy detection technique can be given by the equation
= 1 , , (14)
( 2+ 2 )
where 2 is the variance of the transmitted signal, and
2 is the variance of noise, , is the lower incomplete gamma function given by,
0
0
, = (1)
(15)
and probability of miss detection for energy detection technique is given by,
= 1 , (16)

MATCHED FILTERING METHOD
This method is only applicable when the prior information of the transmitted signal is known, such as the synchronization, modulation techniques, etc [4]. The signal should be synchronized before applying this method. By matched filtering the SNR will be increased and hence give a better detection results. It is best among the three methods to be applied if the transmitted signal information is known to the secondary user (CR) receiver (detector) or in other case if the prior
where is the exponential function, is the received signal power, 2 is the noise variance, and is the threshold value.
Now the probability of detection for the matched filtering detection is given by the alternative hypothesis,
1 which implies the signal is present with noise added with it, and the probability of detection is given by,
= (1 /1 )
= > /1
when is greater than the threshold , now the received signal will be = + ,
Hence the matched filtered signal can be written as,
information is not known, this method cannot be used for
= 1( () + ) Ã— ()
(19)
detection purpose which makes this method poor compared to other two techniques. Also the requirements for generating the copy of the primary signal for each
=0
Now the equation for the probability of detection is given by
different channel require complex hardware adding to its
cost. The block diagram for this method is given Fig. 6.
= 2 , 22 (20)
Sub channel filter
Sub channel filter
1
1
>
>
2
2
= . ()
= . ()
ADC
()
()
E is the received signal power, is the threshold
=0
=0
value, 2 is noise variance, and , is the
generalized Marcum Qfunction, give as
Known information
Known information
2 2
1
1
1
+
, =
2 1
(21)
Fig. 6: Block diagram of Matched Filter Detection technique.
In this figure, the received signal is digitized and the required channel signal is selected from the sub band filtering of (), now this signal is multiplied with the known transmitted signal generated by the secondary user (), the resultant equation () is shown in the block diagram. Now () is compared with the threshold , if > then primary user is using the channel, else the channel is free to be allocated for the secondary user. The test statics for the matched filter detection are given below. The probability of false alarm for the matched filtering is given by the equation
= (1 /0 )
= > /0
and the null hypothesis 0 implies that the received signal will only be noise, given by equation 3. The matched filtered signal is represented by,
where a and b are nonnegative rel numbers, m is positive integer, 1 is modified Bessel function of the first kind of order (m1). Now probability of miss detection for the matched filtering is given by
= (0 /1 ) = 1

CYCLOSTATIONARY DETECTION
In nature all the processes are not periodic functions but their statistical features indicate periodicities and such processes are called cyclostationary process. In this technique, the secondary user can be very easily differentiated between the noise and the signal by analyzing their periodic nature. The noise signal is wide sense stationary and does not have any periodicity, but the TV signals are modulated carrier signals, with cyclic prefixes which have the periodic nature. Hence by estimating the spectral correlation function, noise and signal can be easily differentiated even in very high noise. This makes the method robust compared to other
= 1 Ã— ()
(17)
two methods. Any signal that is wide sense cyclo
=0
Hence the probability of false alarm for matched filtering detection is given by
stationary should have the following criterion: The received signal is () and mean function of the signal is periodic with some period say, 0 given by [5],
= 2 (18)
2
+ 0 = (22)
and its autocorrelation function will be periodic with period 0 .
+ 0 , = (, ) (23) where,
(, ) = + Ã— () (24)
Now the cyclic autocorrelation function (CAF) in terms of Fourier coefficient is given as:
0 = 1 0 /2 , Ã— 2 /0
(25)
0 0 /2
n/0 represents cyclic frequencies, represented by . Now the cyclic spectral density (CSD) also known as the spectral correlation function (SCF) is equivalent to the Fourier transform of cyclic auto correlation function [4]:
Fig. 8: Simulated Spectral Correlation Function, ,
=
=
, =
2
(26)
Probability of false alarm is given by,
In Fig. 7 a signal from the sub channel is selected and it is shifted by the cyclic frequency and the time shifted
= 2(2+1) (27)
2 2
signals are passed through hamming window. The resultant signal is multiplied and we obtain the frequency domain transformed signal: the spectral correlation function (SCF). We estimate the maximum value of the SCF [6],
= max((, )) (27)
and is compared with the threshold , if > then the primary user is present, else the primary user is absent.
where is exponential function, is the number of samples, is the threshold value, 2 is the noise variance. Now the probability of detection for cyclo stationary method is given by:
= (1 /1 )
= (, ) > /1 (28)
= 2 , (2+1) (29)
Sub channel filter
Sub channel filter
()
+
Windowing
FFT
FFT
Windowing
(, )
where is signal to noise ratio, , is Generalized Marcum Qfunction, is standard deviation of noise. Now probability of miss detection for cyclostationary detection is given by
= (0 /1 )
Fig. 7: Block diagram of cyclostationary detection method.
Fig. 8 shows the simulated SCF, in which we find the peaks at specific frequency points = 0, =


RESULTS
= 1
and = 2 , = 0 . The peaks are observed for various
numbers of times and the decision about the channel status is taken [6]. The test statics for the cyclo stationary detection are given below:
The probability of false alarm for the cyclostationary detection is given by the equation
= (1 /0 )
= (, ) > /0
and the null hypothesis 0 implies that the received signal will be only noise, given in equation 3.
We have assumed total number of samples for signal generation as 212 over the entire band. From sub channel filtering, analog TV channel numbered 2 is selected with bandwidth ranging from 5460 MHz, and video carrier at
54.25 MHz, and audio carrier at 59.75 MHz. The signal obtained contains N=125 discrete samples. This signal is tested with all the three detection techniques and probability of false alarm( ), probability of detection ( ), probability of miss detection ( ) for all the three methods viz. energy detection, matched filter detection, and cyclostationary detection techniques are compared below for SNR ranging from 40 to 40dB.
Probability of false alarm( ): In Fig. 9 the probability of false alarm for cyclostationary detection is 1 up to
25 dB and goes on decreasing as SNR increases, and
reaches zero at nearly 5 dB SNR. At this stage values for matched filtering and energy detection are still 1. Similarly of matched filter decreases at 3dB SNR and reaches zero at 19 dB SNR; for energy detection decreases from 1 at 28 dB SNR and reaches zero at 32dB SNR. Here it shows that cyclostationary outperforms matched filtering and energy detection, and the matched filtering is better than the energy detection.
Probability of detection( ): In Fig. 10 the Probability of false detection for cyclostationary detection is 0 upto 35dB and goes on increasing as SNR increases, and reaches 1 at nearly 32dB SNR. At this stage for matched filtering and energy detection are still zero. Similarly of energy detection increases from zero at – 18dB SNR and reaches 1 at 9dB. For matched filtering
increases from 0 at 15dB SNR and reaches 1 at 5dB. Again, it shows that cyclostationary outperforms the matched filtering and the energy detection; and the energy detection is better over the matched filtering.
Fig. 9: Simulated probability of False Alarm.
Fig. 10: Simulated probability of Detection.
Fig. 11: Simulated probability of Miss Detection.
Probability of miss detection ( ) : In Fig. 11 the simulated probability of miss detection for cyclo stationary detection is 1 up to 35dB and goes on decreasing as SNR increases, and reaches zero at nearly – 32dB SNR. At this stage values for matched filtering and energy detection are still 1. Similarly of energy detection decreases at 18dB SNR and reaches zero at – 9dB SNR. For matched filtering decreases from 1 at – 15dB SNR and reaches zero at 5dB SNR. Here it shows that cyclostationary technique outperforms the matched filtering and the energy detection technique. The Energy detection technique is better than the matched filtering technique.

CONCLUSION
In this paper we have compared the performance of
energy detection, matched filter detection and cyclo stationary detection techniques with respect to their probabilities of false alarm, detection and miss detection. It is observed that energy detection does not require prior knowledge of the transmitted signal, makes it better with respect to matched filtering, but for very low SNR, it fails to give correct results. If the transmitted signal characteristics are known then matched filter technique is better than energy detection. But it is more costly for its complex hardware requirements compared to energy detection. This comparative study shows the cyclostationary detection technique outperforms both the energy detection and matched filter detection due to its capability to differentiate the signal from the noise. It is suited well for low SNR condition, compared to energy detection and it does not require any prior information about the transmitted signal as in the case of matched filtering technique. The superiority of cyclostationary technique comes at the cost of increased computational time and complexity.
ACKNOWLEDGEMENT
Authors would like to thank Mr. Ravindra Noubade, director and Mr. Jaideep Bose senior software engineer nd consultant, both from Asmaitha Wireless Technologies Pvt. Ltd, Bangalore, for their support in doing this project work. Authors also acknowledge management, Dayananda Sagar Group of Institutions (DSI), Bangalore for all its support in pursuing this research in Research center, Department of Telecommunication Engineering, Dayananda Sagar College of Engineering, Bangalore, India. Our special thanks are due to Dr. Premachandra Sagar, Vice chairman, DSI for all his encouragement for research.
REFERENCES

Vicente Osa, Carlos Herranz, Jose F Monserrat and Xavier Gelabert, Full text Implementing opportunistic spectrum access in LTEAdvanced, EURASIP journal on Wireless communication and Networking, March 2012, 118.

IEEE 802.22 Working Group on Wireless Regional Area Networks website http://www.ieee802.org/22.

D. Wyatt A. Taylor III, Design of Ultra wideband Digitizing Receivers for the VHF Low Band, Virginia Polytechnic Institute and State University, May 9th, 2006, 180.

Ashish Bagwari, Brahmjit Singh, Comparative performance evaluation of Spectrum Sensing Techniques for Cognitive Radio Networks, in proc. of Fourth International Conference on Computational Intelligence and Communication Networks, 2012 98105.

Alfateh M. Mossa, Varun Jeoti, Cognitive Radio: CyclostationarityBased Classification Approach for Analog TV and Wireless Microphone Signals, in proc. of international conference on Innovative Technologies in Intelligent Systems and Industrial Applications (CITISIA 2009) Monash University, Sunway campus, Malaysia. 25th & 26th July 2009, 107111.

Manish B Dave, Spectrum Sensing in Cognitive radio: Use of CycloStationary Detector, NATIONAL INSTITUTE OF TECHNOLOGY Rourkela, Orissa, India. May 2012, 1723.