A Novel Approach for Cognitive Radio Sensing using Wideband Chirp Signal

Ahmed Barnawi
Faculty of Computing and IT, King Abdul-Aziz University, P.O. Box 80221, Jeddah 21589, Saudi Arabia

ambarnawi@kau.edu.sa
Abstract sensing the radio environment is probably the most important and challenging aspect in cognitive radio systems. Traditional sensing techniques have proven shortcomings ranging from inaccuracy to implementation complexity. Chirp signal is a wideband signal with variety of applications in communication and signal processing. In this paper, we introduce a novel approach for sensing the radio spectrum using chirp signal. This method is showing potential to improve sensing in cognitive radio environment at tolerable level of complexity. Index terms - cognitive radio, radio sensing, chirp signal.

its matched filters compressed output. The use of chirp signal will ease the complexity and improve the quality of sensing and therefore offer enhanced cognition at the CR user. This paper is organized as follows: In II, we briefly talk about the chirp signal and its characteristics. In III, we outline the network architecture. In IV, we explain our sensing methodology. In V, mathematical analysis is briefly explained. In VI, we present the simulation model. In VII, the results are shown and discussed. In VIII we conclude. II. CHIRP SIGNAL

I.

INTRODUCTION

Cognitive Radio (CR) Systems is attracting a lot of attention due to the scarcity of radio resources. These systems benefit from low utilization of the radio frequencies in licensed frequency bands. Thus CR users are opportunistic users of the radio channel in the way that they try to place their transmissions in unutilized frequency gaps without causing excessive interference to the transmission of the primary users of the radio channel- primary user is a user licensed to use the channel. The success of this transmission strategy will relay on the quality and quantity of cognition possessed at CR user. This cognition is acquired through rigorous sensing of the radio channel and ability to characterize the interference. Based on the sensing functionality, CR users are ought to adapt there transmission accordingly. The problem of spectrum sensing and characterization is a typical trade-off problem where the accuracy and system simplicity are inversely related. The most known sensing techniques used are match filtering [1], energy detection [1], and cyclostationary features detection [2]. Match filtering is the optimal detection however due to system requirements it is practically difficult to implement [1]. Though at lower level of difficulty the performance of cyclostationary features detection is near optimal, system complexity is not trivial [1]. Energy detection is the least complex and most inaccurate among the three methods [3]. In this paper we introduce a novel sensing technique in an infrastructural cognitive radio network architecture that is based on the use of the chirp signal. Chirp signal is a wide band signal that has interesting characteristics, particularly

Wideband chirp modulation is a result of linear frequency modulation for digital signal. The instantaneous frequency of the chirp signal increases or decreases linearly with the time, Figure 1a shows a chirp signal. The bandwidth of a chirp signal, F, extends from the starting frequency sweep, f1, to the final frequency sweep f2. With proper choice for processing gain i.e. FT product, where T is the bit period, the spectrum of chirp signal has a distinctive near square shape, Figure 1b. (a)

(b)

F
0 1 2 3 Frequency (kHz) 4

Figure 1: (a) Chirp signal, (b) Chirp spectrum

978-1-4244-5753-3/10/$26.00 2010 IEEE

III.

SYSTEM ARCHITECTURE

Figure 2 shows the system architecture architecture. The system is hybrid and contains two networks; a primary mary radio network and a cognitive adaptive radio network [4]. The two networks are not physically connected however they meant to coexist. A. Primary Radio network The primary radio network is the essential part of the system. It consists of a primary base station serving primary licensed users over the primary coverage are area. The primary base station performs normal functions of a base station. B. Cognitive Radio network The cognitive radio network is the adaptive part of the system. It consists of cognitive radio base station (CR (CR-BS) serving cognitive radio users (CR-user). user). The cognitive radio coverage area overlaps with primary coverage area. A CR CRuser is ought to behave ehave in opportunistic fashion where it only can transmit after sensing the availability of the radio resources thus guarantee no excessive interference occurs at a primary users receiver.

to decide the minimum amplitude of utilized frequencies. A. Deciscion Circuit Decision circuit is an algorithm implemented by software to detect the peaks representing primary users carrier frequency. This algorithm could be implemented sequentially or binary. Sequential algorithm will sweep across the magnitude values of FFT samples sequentially comparing them to a threshold value. However the binary algorithm will first sort the values of the FFT samples then it will discard the samples below low the threshold value. both algorithms will return the frequency values at which the FTT magnitude values exceed the threshold. V. ANALYSIS

Mathematical modeling can prove the concept of our methodology. Figure 3a shows the block diagram of the system. The inputs to the chirp signal matched filter is the referenced chirp signal f(t), and the interfering carrier signals of primary users m1(t), m2(t) For simplicity, we will address the problem in the frequency domain as time convolution convol is transferred into frequency multiplication, Figure 3b. . Therefore: F )M2() G() = F2()+ F() M1() + F( (1)

Since the frequency conversion for a sine wave is the unit impulse function in the frequency domain shifted to its corresponding frequency, (2) can be further simplified as: G() = F2()+ M1() + M2() (2)

Assuming that the spectrum trum of chirp signal is a square wave function, Figure 3c visualizes G().
Figure 2: system architecture

IV.

SENSING METHODOLOGY

. f(t), m1(t), m2(t)

h(t) = f(-t) Matched filter (a)

g(t)=h(t)*f(t) g(t) + h(t)*m1(t) + h(t)*m2(t)

The novel sensing methodology could be summarized as follow: 1. CR-BS BS broadcast a low power reference chirp signal with a bandwidth covering the sensed spectrum. 2. After traveling over the radio channel and interfering with primary users' transmissions ons and noise, the reference chirp signal is then received at the CR CR-user using a chirp signal matched filter. 3. Fast Fourier Transform (FFT) is then applied to the output of the matched filter. Figure 5 shows the spectrum of the received reference signal. As it is shown, the utilized (interfering) frequencies appears as spikes (peaks) rising above the flat top of the received chirp signal. 4. The output from the previous step is fed into a decision circuit stage where a threshold value is set

F() F*() M1(), M2() Matched filter (b) F2() + 1 4

G() = F()2 +M1()+ M 2( )

G() M 1( ) 2 (c) Figure 3: Analysis + M 2( ) 3 =

VI.

SIMULATION MODEL

Figure 4 shows a block diagram for the proposed algorithm implemented using Matlab. The reference chirp signal is received at the CR-receiver after interfering with primary users signals in AWGN channel. The Chirp signal is firstly received by chirp signal matched filter. Then FFT is applied and the output of FFT passes to the decision circuit to decide whether primary users interfering with the referenced chirp signal.

fine tuned by CR-user to minimize probability of false detection.

(a)
peak

Chirp spectrum floor

AWGN
Chirp (Ref) Chirp MF Decision Circuit

+
Primary user

FFT
CR receiver

Primary user

Primary user

(b)
Figure 4: simulation model

VII. RESULTS AND DISCUSSION Figure 5 shows spectrum of received chirp signal. The interesting characteristic is obvious where a near flat floor extend over the bandwidth i.e. 3600 Hz. For chirp signal period of 1 s, the processing gain is 3600. It is shown that two distinctive peaks occur at frequencies corresponding to primary users carrier frequencies at 500 Hz and 1800 Hz. Figure 5a shows the scenario where Signal to Interference plus Noise Ratio (SINR) for the interfering (primary users) carriers is 20 dB. Figure 5b shows the scenario where SINR is -5 dB. It is obvious that as SINR decreases, noise floor rises toward the peak value. In order to quantify the performance of the system we define d which measures the distance (in dB) between the peak of the carriers spike and the flat floor of chirp signal spectrum. In Figure 6, we plot the carriers SINR versus d, normalized to value of d at SINR = 10 dB. The value of d = 0 dB signifies that the spike is no longer distinguishable from the noise and therefore probability of false alarm is very likely. The level of the noisy floor should determine the threshold value for the decision circuit. It is obvious that as SINR decreases d decrees. For our setup, it is shown that d = 0 at SINR= -25 dB which is extremely a low SINR value. From above discussion it is becoming important for a cognitive user to be able to set the threshold value for its decision circuit. This process which depends on the level of SINR of received reference chip signal should be dynamic as the value of SINR changes upon many factors in the mobile environment. In order to address this problem, CR-user must be capable of estimating the SINR of received reference signal. Also The CR-BS must send a testing tone which is going to be used by CR-user to estimate the value of d based upon prior knowledge of system receivers behavior at given SINR as shown in Figure 6. This estimation could be further

Figure 5: received chirp spectrum (a) SINR = 20 dB, (b) SINR = -5dB

1.2 1 0.8 0.6 0.4 0.2 0 -30 -20 -10 0 10 20

Figure 6: SINR versus normalized d.

The results in Figure 6 assume that primary users transmissions are in synch with the reference chirp signal. The performance is expected to degrade as this assumption is

relaxed. Figure 7 illustrates the delay scenario between reference chirp signal and primary users carrier.

Tc delay Tb
Figure 7: delay between reference chirp signal and received primary signal

To investigate the effect of a more realistic scenario where both networks (primary and cognitive) are not synchronized, Figure 8 shows how d changes with respect to the delay between primary users signal and reference chirp signal. The results shows SINR in two cases, SINR = 10 dB and -10dB. As expected, the system degrades as delay increases. For example, if the users signal is delayed by 0.5Tc s, Tc is the chirp signal bit period, d drops by 25% in both SINR scenarios. Nonetheless, the system is showing tolerant to delay up to 0.25Tc s in both SINR cases. Also from Figure 8, we can deduce that below -10 dB the performance will greatly degrades especially if synchronization is not maintained. 1.2 1 0.8 0.6 0.4 0.2 0 0 0.2 Tc 0.4 Tc 0.6 Tc 0.8 Tc
SINR = -10 dB SINR = 10 dB

duration, the sensing strategy will fail. To drive the relation between Tb and Tc, we need to observe the relation between the delay and d in Figure 8. It is shown that in worst case scenario i.e. SINR = -10 dB, d has values greater than 0 if the delay is less than 0.75Tc. This is interpreted as that the minimum overlapping duration should be greater than one forth the chirp signal duration. On the other hand if we want to estimate the duration of overlapping which is enough to give optimal performance, we need to observe the performance between delays equal to 0 to 0.25Tc. It is shown that between delays equal to 0 to 0.25Tc, d remains unchanged regardless to the SINR. This is interpreted as that it is enough for Tb to last 0.75Tc. To summarize this point, we imply that 0.25Tc < Tb < 0.75Tb or 1.33Tb < Tc < 4Tb. The above discussion entitles us to highlight how our strategy addresses the main shortcoming of other known sensing strategies [4] i.e. energy detection, matched filtering and cyclostationary feature detection. As for energy detection, the main problem is setting the threshold value in presence of background noise. This problem is alleviated by the virtue of flat top of chirp signal matched filter output which maintains good resolution even at low SINR. As for match filtering, the main problem is that this method requires priori knowledge of primary user signal for optimal detection. In our strategy, this is not the case as priori knowledge of primary user is not necessary for optimal detection. Finally, in cyclostationary feature detection, the main problem is increasing complexity however it is shown that the complexity of our system is moderated. VIII. CONCLUSION

Our tested novel method for sensing cognitive radio environment significantly enhances sensing at low complexity. It can sense transmissions of up to -25 dB however it is shown to be sensitive to synchronization between reference chirp signal and primary users especially at low SINR. This sensitivity could be overcome if observation interval increases. Future work should evaluate the strategy in mobile environment. ACKNOWLEDGEMENT The author wishes to acknowledge Prof. Abraham Fapojuwo from University of Calgary for his support and assistance. REFERENCES
D. Cabric, S.M. Mishra, R. W. Brodersen, Implementation Issues in Spectrum Sensing for Cognitive Radios, 38th Annual Asilomar Conference on Signals, Systems and Computers, November 2004. [2] W. A. Gardner, C. M. Spooner, Signal Interception: Performance advantages of Cyclic-feature detectors, IEEE Transaction on Communications, vol. 40, January 1992. [3] Tevfik Yucek and Huseyin Arslan, Spectrum Characterization for Opportunistic Cognitive Radio Systems, IEEE MILCOM 2006, Octorber 2006. [4] Kwang-Cheng Chen and Ramjee Prasad , Cognitive Radio Netwoks, Page 183-186, John Wiley & Sons Ltd, 2009.
[1]

Figure 8: delay versus normalized d.

It must be noted that the results shown in Figure 8 simulate the scenario of one chirp signal with bit period Tc and a delayed carrier with bit duration of Tb. Also it must be noted that in this simulation Tb = Tc. The obvious conclusion to be drawn from this is that d is dependent on the overlapping duration between the chirp duration and the bit duration. Therefore, even in worst case scenarios when the overlapping between the first chirp duration and the carrier bit is not enough, the consecutive overlapping should be enough to make a better decision. This may lead to the conclusion that observation interval must be at least twice as much as chirp signal period to avoid this drawback. From this discussion, we can determine other important criteria necessary for the success of the sensing strategy which is the relation between Tb and Tc since it is evident that if Tb is not long enough to produce enough overlapping