SPECTRUM SENSING USING BUSSGANG THEOREM FOR IEEE 802.

22 WRAN
Sung Sue Hwang, Dong Chan Park, Suk Chan Kim, Myung Sun Song*, and Sun Min Lim* Department of Electronics Engineering, Pusan National University Geumjeong-gu, Busan, 609-735, Korea *Electronics and Telecommunications Research Institute *138 Gajeongno, Yuseong-gu, Daejeon, 205-700, Korea sckim@pusan.ac.kr ABSTRACT Utilization problem of the limited spectrum is the one of the most important issues in wireless communication systems. Cognitive radio technique which is finding and utilizing frequency holes is also one of those techniques. Specially, the spectrum sensing technique to detect the primary user signal is a core technology in cognitive radio area. In this paper, we propose the spectrum sensing algorithm using Bussgang theorem. The proposed algorithm calculates the statistical difference between the Gaussian noise and the primary user signal by applying Bussgang theorem to the received signal. The algorithm is not affected by noise uncertainty and can detect the primary user signal in the very low SNR (signal-to-noise power ratio) environment. We evaluate the algorithm through computer simulations with 12 ATSC A/74 DTV signal captures based on IEEE 802.22 WRAN (wireless regional area network) I. INTRODUCTION Most of the usable spectrum is already allocated to conventional systems, but the allocated spectrum is not efficiently used. Much of the allocated spectrum is unused in the certain time and area. The FCC (federal communication commission) reported that there are temporal and geographic variations in the usage of allocated spectrum ranging from 15% to 85% [1]. In 2004, the IEEE formed 802.22 WG (working group) to establish an international standard for WRAN (wireless regional area networks) based on CR (cognitive radio) [2].1 Basic concept of the WRAN system is that WRAN system detects primary user, identifies idle channels, exploits these channels for a broadband wireless communication, and avoid primary users. For this basic concept, spectrum sensing is one of the most important components of the WRAN system. Sensing Tiger Team of the WRAN WG for the evaluation of proposed sensing
This work was supported by the IT R&D program of MKE/IITA [2005-S-002-04, Development of cognitive radio technology for efficient spectrum utilization] 978-1-4244-2677-5/08/$25.00 2008 IEEE 1 of 5

algorithms require an accurate sensing method satisfying detection of primary user signal when the SNR (signal to noise ratio) is below -20dB. Some sensing algorithms including the energy detection [3], cyclostationary detection [4], covariance based detection [5], maximum-minimum eigenvalue detection [6], and signature based detection [7] were proposed to the WRAN. These methods have different advantages and disadvantages. Energy detection is simple and detects signals without prior knowledge of signals. But sensing performance of energy detection method is insufficient in very low SNR and sensitive to noise uncertainty [8]. Cyclostationary detection performs well when SNR is very low, but there is performance degradation due to noise uncertainty. Signature based detection utilizes the sync signal of the ATSC (advanced television systems committee) DTV (Digital Television) signal, so that it can identify ATSC DTV signal, but does not work in very low SNR. Covariance based detection and maximum-minimum eigenvalue detection are not related to noise uncertainty, but these methods cannot detect signal in very low SNR. We propose spectrum sensing algorithm using Bussgang theorem, and the proposed algorithm is not related to noise uncertainty anymore and can meet the requirement of the WRAN. The proposed algorithm calculates the statistical difference between the Gaussian noise and the primary user signal by applying Bussgang theorem [9]. If the received signal includes the non-gaussian primary user signal, the proportionality of auto-correlation to crosscorrelation is different from the proportional constant of the Gaussian signal (the Gaussian noise). Bussgang gaussianity test for stationary series use this approach [10]. Unfortunately, ATSC DTV signal as a primary user signal for WRAN follows the Gaussian distribution in the time domain, so that the proposed method cannot detect ATSC DTV signal in time domain. But ATSC DTV signal does not follow the Gaussian distribution in the frequency domain due to the pilot tone, therefore we can apply proposed sensing method to detect the signal. This paper is organized as follows. We present the spectrum sensing system model in section . In section

, we present a brief description of Bussgang theorem and the spectrum sensing using Bussgang theorem. We describe the proposed sensing algorithm for IEEE 802.22 WRAN, and theoretical analysis of test statistic and threshold in section . Simulation results demonstrating the sensing performance of the proposed method in the low SNR is presented in section . Finally, conclusion is given in section . II. SPECTRUM SENSING SYSTEM MODEL Assume that there is a single sensor with M receivers (antennas). The aim of the detection problem is to distinguish between the two hypotheses (H0, H1). The received signal ri[n] through the i-th receiver is described in the following: H0: ri [n] = wi[n] H1: ri [n] = s[n]*hi[n] + wi[n] (1) (2)

variance Gaussian stationary process, the proportional constant Kg is described as

Kg =

1 2

x g ( x )e

x2 /2

dx

(4)

Kg is a function of g(). If the signal to be detected is not a Gaussian stationary process, there is some difference between the theoretical proportional constant and the sample proportional constant. Base on this difference, we can decide whether or not the objective signal exists. The sample proportional constant Ki[] at the receiver i is given by

K i [ ] =

r [n] g (r [n + ])
i n=0 N 1 n=0 i

N 1

r [ n] r [ n + ]
i i

(5)

where s[n] is the transmitted signal from a primary user, hi[n] is the channel impulse response from the primary user to the i-th receiver, and wi[n] is the white Gaussian noise (i=1,2,M). The hypothesis H0 assumes that the primary user signal is absent hence only the white Gaussian noise is present in ri[n]. The hypothesis H1 assumes that the primary user signal s[n] to be detected is present. The performance measures of a sensing algorithm are the probability of missed detection (PMD), and the probability of false alarm (PFA). PMD is the probability that the sensing algorithm is unable to detect the presence of the primary signal at the hypothesis H1. PFA is the probability that the sensing algorithm decides that the signal s[n] exists at the hypothesis H0. The good sensing algorithm requires low PMD and PFA in the low SNR, however there is a trade-off between PFA and PMD. III. BUSSGANG THEOREM AND SPECTRUM SENSING USING BUSSGANG THEOREM Bussgang theorem states that when a real Gaussian stationary process passes through a non-memory nonlinear device, the cross-correlation function of input and output is proportional to the auto-correlation function of input [9], that is,

where ri [ n ] is the normalized version (zero mean, unitvariance) of the received signal at the receiver i, is a time delay(-N,,N), and N is the number of the received signal sample. The test statistic is defined as

T [ ] = ( K i [ ] K g )
M 1 i=0

(6)

If T[] > , the signal is present, while if T[] , the signal is absent where is the sensing threshold. For hypothesis H0, there is only the Gaussian noise in the received signal and we can derive the theoretical threshold is decided by the required PFA. If is zero, the proportional constants are equal to the cross-correlations of the signal and the distorted signal, and follow the Gaussian distribution with the mean Kg, the standard deviation g based on Central Limit Theorem. The T[0] follows the Morder chi-squared distribution. The cumulative distribution function of T[0] is in the following

FT ( y ) =

M 0 g

1 2 M / 2 ( M / 2)

u M / 2 1e

2 u / 2 g

du

(7)

Kg =

E{x[n] g ( x[n + k ])} E{x[n] x[n + k ]}

(3)

where () is a gamma function. For the required PFA, the corresponding threshold is FT() = 1 - PFA. Finally, we obtain the threshold for the proposed algorithm as = FT1 (1 PFA ) (8) The threshold is determined by required PFA and used nonmemory non-linear function.

where x[n] is a real Gaussian stationary process, g() is a non-memory non-linear function and Kg is the proportional constant. When the input x[n] is the zero-mean unit-

2 of 5

r[n]
BPF

e j 2 fst /8
LPF, Down Sampling

Kg

Threshold

x[n]
FFT

X[k]

Normali zation

YRr[k] YIi[k] ]

Calculate

] K K ] r[k R

K [ ]

K KIi[k ]

Test statistic

T[]
Decision

Figure 1. Block diagram of spectrum sensing using Bussgang theorem for IEEE 802.22 WRAN

IV. SPECTRUM SENSING FOR IEEE 802.22 WRAN In this section, we describe spectrum sensing method to detect the ATSC DTV signal. There are several primary user signals for the IEEE 802.22 WRAN, but we consider only the DTV signal as a primary user signal. This signal is 8VSB (vestigial sideband modulation) modulated with the sampling frequency fs = 21.5244MHz, and down converted to 5.38MHz, hence the pilot tone of the signal is close to 2.69MHz. Figure.1 illustrates the block diagram of the proposed spectrum sensing algorithm. We assume there is a single sensor with a single receiver. The received signal is filtered with bandwidth 6MHz, and down converted to place the pilot tone at the baseband. The ATSC DTV signal is a Gaussian process in the time domain, so the sensing algorithm cannot detect the time domain DTV signal in the previous section. However, the frequency domain DTV signal is not a Gaussian process because of the pilot tone. Consequently, we filter the DTV signal by narrowband baseband filter with the bandwidth BW = NFFT / TZ to obtain the pilot tone. The filtered pilot tone is decimated by a factor of floor( fs / BW), and the down sampled signal x[n] is converted to the frequency domain using FFT(fast Fourier transform) of length NFFT. T = 5ms sensing time, and Z is a sensing time constant. If Z = 1, 2, 3, .., the total sensing time TZ is 5, 10, 15, ms. In the case of H0, the frequency domain signal X[k] is a white Gaussian process. We divide X[k] into real and imaginary part, and normalize the each part for zero-mean unitvariance. The normalized signal YR[k] and YI[k] are given by

T [ ] = ( K R [ ] K g ) 2 + ( K I [ ] K g ) 2 .

(10)

The test statistic T[0] is the sum of two zero-mean Gaussian random variable, therefore T[0] follows the 2ndorder chi-squared distribution. The cumulative distribution function of T[0] is given by

FT ( x) = 1 e

2 x / 2 g

(11)

where g is the standard deviation of the KR[0], and KI[0], it can be easily derived with the property of sum-process. Finally, we obtain the theoretical threshold for the proposed algorithm as
2 = 2 g ln( PFA ).

(12)

The threshold is not related to the sensing time, and the noise variance, but the threshold vary with g. V. SIMULATION RESULTS In this section, the sensing performance of the proposed algorithm to detect ATSC DTV signal is considered. The initial signal processing of the signal is based on [11]. We follow the simulation steps and the WRAN spectrum sensing scenario 2 (a single WRAN spectrum sensor with a single receiver) in [12]. Table.1 describes used ATSC A/74 captured DTV files in our simulations. The used parameters are NFFT = 2048, = 0, fs = 21.5244MHz, T = 5ms, Z = 1, 2, 4, BW = NFFT / TZ, and PFA = 0.1. Figure 2, 3, and 4 show average sensing performances of proposed algorithm over 12 DTV files, and the sensing performance of MME (the maximum-minimum eigenvalue detection) as a function of the SNR at the total sensing time 5, 10, and 20ms. The size of the covariance matrix is 10 by 10 in MME. At the sensing time 5, 10, and 20ms, the proposed algorithm achieve PMD = 0.1 subject to PFA = 0.1 when SNR = -17, -20, and -22 dB. As can be seen a longer sensing time results better performance due to the higher resolution of the pilot tone and SNR of the narrow base-band filter. The sensing performance is also related to

YR [k ] =

R I X R [k ] m X [k ] m , YI [k ] = I . R I

(9)

where XR[k] is the real part of X[k], XI[k] is the imaginary R and m I are the estimated means of XR[k], part of X[k], m

R and I are the estimated standard XI[k], and deviations of XR[k], XI[k]. We calculate the each proportional constant KR[] and KI[] using (5). Finally, the test statistic T[] is defined as

3 of 5

Table 1. ATSC DTV Captured Data Files

ATSC DTV Captured Data File WAS_68_36_05232000_ref WAS_86_48_07122000_ref WAS_51_35_05242000_REF WAS_49_39_06142000_opt WAS_32_48_06012000_OPT WAS_3_27_06022000_REF WAS_47_48_06132000_opt WAS_311_ 48_06052000_ref WAS_311_35_06052000_ref WAS_06_34_06092000_ref WAS_49_34_06142000_opt WAS_311_36_06052000_ref Average over 12 DTV files legend
1 0.9 0.8 0.7 0.6

MD

0.5 0.4 x3 0.3 0.2 0.1 0 -34 x5 x7 sgn(x)

2 sgn(x)(1-e-x /2) eig

-32

-30

-28

-26

-24

-22

-20

-18

SNR(dB)

Figure 3. Average sensing performance (10ms)

1 0.9 0.8 0.7 0.6

1 0.9 0.8 0.7 0.6

PMD

MD

0.5 0.4 x3 0.3 0.2 0.1 0 -30 x

5 7

0.5 0.4 0.3 0.2 0.1 0 -38 x3 x5 x7 sgn(x)

2 sgn(x)(1-e-x /2) eig

x sgn(x)
2 sgn(x)(1-e-x /2) eigen

-28

-26

-24

-22

-20

-18

-16

-14

-36

-34

-32

-30

-28

-26

-24

-22

-20

-18

SNR(dB)

SNR(dB)

Figure 2. Average sensing performance (5ms) g(). Especially, power series functions show good performances among several non-linear functions. Figure.5 demonstrated that when the higher and odd order function is used, the proposed algorithm achieves higher sensing performance. However, the complexity of the propose algorithm is affected by g(), therefore we should select the proper g() in considering the required sensing performance and the computational complexity. Figure.6 depicts performance of each 12 DTV files when g(x) = sgn(x)[1-exp(-x2/2)]. There are significant differences among 12 DTV signals, because the pilot tones of 12 DTV signals have different frequency offset and power.
VI. CONCLUSION

Figure 4. Average sensing performance (20ms) between the Gaussian noise and the primary user signal by applying Bussgang theorem to the received signal. ATSC DTV signal can be detected in frequency domain due to it follows the Gaussian distribution in time domain. Simulation results demonstrate that proposed algorithm can achieve PMD = 0.1 subject to PFA = 0.1 when the sensing time = 20 ms and SNR = -22dB. The non-memory non-linear function determines the sensing performance and the complexity therefore it should be selected after considering required sensing performance and computational complexity. Though the most of the conventional sensing algorithm is affected by the noise uncertainty, the sensing performance of the proposed algorithm is not related to noise uncertainty.
REFERENCES

In this paper, we proposed the spectrum sensing algorithm using Bussgang theorem for IEEE 802.22 WRAN. In addition we discussed the theoretical analysis of the test statistic and the threshold. Proposed algorithm detects the objective signal using the statistical difference
4 of 5

[1] FCC, Spectrum Policy Task Force Report, ET Docket No. 02-155, Nov. 2002.

0.5 0.45 0.4 0.35 0.3 g(x) = x 3 g(x) = x 5 g(x) = x 7 g(x) = x 9 g(x) = x 2 g(x) = x
4

1 0.9 0.8 0.7 0.6

PMD

0.25 0.2 0.15 0.1 0.05 0 -26

PMD
-18

g(x) = x 6 g(x) = x 8

0.5 0.4 0.3 0.2 0.1

-25

-24

-23

-22

-21

-20

-19

0 -38

-36

-34

-32

-30

-28

-26

-24

-22

-20

-18

SNR(dB)

SNR(dB)

Figure 5. Average sensing performance, g(x) = xN (20ms) [2] J. Mitola III, Cognitive Radio: An Integrated Agent Architecture for Software Defined Radio, Ph.D. Thesis, Royal Institute of Technology, Sweden, May 2000. [3] S. Shellhammer, Text on energy detector, IEEE 802.22-07/0264r2, Jun. 2007. [4] Hou-Shin Chen, Wen Gao and David G.Daut, Spectrum sensing using cyclostationary properties and application to IEEE 80.22 WRAN, in Proc. GLOBECOM 07, pp. 3133-3138, Nov. 2007. [5] Y. Zeng and Y. Liang, Covariance based signal detections for cognitive radio, in Proc. IEEE DySpan07, pp. 202-207, Apr. 2007. [6] Y. Zeng and Y. Liang, Maximum-minimum eigenvalue detection for cognitive radio, in Proc. IEEE PIMRC 07, pp. 1-5, Sept. 2007. [7] H.-S. Chen, W. Gao and D. G. Daut, Signature based spectrum sensing algorithms for IEEE 802.22 WRAN, IEEE ICC, pp.6487-6492, Jun. 2007. [8] S. Shellhammer and R. Tandra, Performance of the power detector with noise uncertainty, IEEE 802.2206/0134r0, Jul. 2006. [9] J.J. Bussgang, Cross-correlation function of amplitude-distorted gaussian signals, Res. Lab. Elec., Mas. Inst. Technol., Cambridge MA, Tech. Rep. 216, Mar. 1952. [10] G. Giunta, G. Jacovitti, and G. Scarano, Bussgang non-gaussianity test for stationary series in Proc. IEEE Signal Processing Workshop on Higher-Order Statistics, pp.21-23, Jul. 1997. [11] S. Mathur, R. Tandra, S. Shellhammer and M. Ghosh, Initial signal processing of captured DTV Signals for Evaluation of Detection Algorithms, IEEE 802.22-06/0158r5, Sept. 2006. [12] S. Shellhammer, V. Tawil, G. Chouinard, M. Muterspaugh and M. Ghosh, Spectrum Sensing

Figure 6. Sensing performances of 12 DTV files, g(x) = sgn(x){1-exp(-x2/2)}(20ms) Simulation Model, IEEE 802.22-06/0028r10, Sept. 2006.

5 of 5