L OS MIGRANTES LIMÍTROFES EN LA CONSTRUCCIÓN DE LA OTREDAD

One of the major challenges of cognitive radio networks is the development of efficient spectrum sensing techniques for the SUs. Spectrum sensing refers to the phase during which the SUs must sense the radio frequen- cies in order to make a decision on whether to transmit or not, depending on the state of the PUs. The main objective of spectrum sensing is the de- sign of high quality spectrum sensing devices and algorithms for exchang- ing spectrum sensing data between nodes to (i)- Reliably detect spectral holes for use by the cognitive radio devices and (ii)- Reliably detect when the primary transmitter becomes active. In order to achieve those goals,

Spectrum Sensing in Cognitive Radio Networks

cross-layer design problems need to be addressed by exploiting advanced digital signal processing techniques, introducing efficient detection and estimation approaches as well as exploiting users’ cooperation.

By using local measurements and local observations, a secondary user can detect the transmitted signal from a PU. The model for signal detection at time t can be described as follows [19, 54]:

y(t) =

{

n(t), H0,

h· s(t) + n(t), H1,

(1.9)

where y(t) is the received signal at an SU, s(t) is the transmitted signal of the licensed PU, n(t) is the additive white Gaussian noise (AWGN), and

h is the channel gain. In (1.9), H0 and H1 represent, respectively, the

hypotheses of having and not having a signal from a licensed PU in the target frequency band. Consequently, the spectrum sensing phase boils down to a decision between two hypotheses H0 or H1, depending on the

received signal at the SU. In order to detect the signal of the PU, different methods can be used such as

1. Matched Filter Detection: Matched filter detection is generally used to detect a signal by comparison between a known signal (i.e., template) and the input signal. It is well known that the optimal method for signal detection is through a matched filter [49], since it maximizes the received signal-to-noise ratio. In addition, by using a matched filter detector, the detection of the PU signal can take a small amount of time [54] which is one of the main advantages of matched filter detection. However, utilizing a matched filter in spectrum sensing requires demodulation of a PU signal which implies that the cogni- tive radio must have a priori knowledge of different PHY and MAC characteristics of the PU signal such as pulse shaping, packet for- mat, and so on. Further, in the event where this information is not available or is incorrect, the performance of spectrum sensing de- grades significantly [19, 54]. As a result, matched filter detection is mainly useful whenever the PU can convey some information on its signal using some sort of pilot channel, preambles, spreading codes, or other techniques that can help the SUs to construct an estimate of the signal.

2. Energy Detection: While the matched filter approach requires coher- ent detection, a more simplified filtering approach is to perform non-

coherent detection through energy detection. Whenever the informa- tion on the PU signal is unavailable at the SUs, energy detection can be quite a useful approach [19, 54]. For energy detection, the out- put signal from a bandpass filter is squared and integrated over the observation interval. Subsequently, a decision algorithm compares the integrator output with a threshold to decide whether a licensed user exists or not [19, 54]. Basically, for energy detection, the perfor- mance deteriorates as the received SNR from the PU signal decreases. Energy detection has been widely adopted in many spectrum sensing scenarios [19]. Despite its practicality and appeal, energy detection suffers from three main drawbacks. First, it is susceptible to the uncertainty of noise. Second, energy detection can only detect the presence of the signal without being able to differentiate the type of the signal. As a result, energy detection can confuse signals result- ing, for example, from other SUs with the PU signal. In addition, an energy detectors do not work for spread spectrum signals, for which more sophisticated signal processing algorithms need to be devised. 3. Cyclostationary Feature Detection: The transmitted signal from a li-

censed PU generally possesses a period pattern. Such signals are commonly referred to as cyclostationary. By using this period pat- tern one can detect the presence of a licensed PU [54]. A signal is cyclostationary (in the wide sense) if the autocorrelation is a periodic function. With this periodic pattern, the transmitted signal from a licensed PU can be distinguished from noise which is a wide-sense stationary signal without correlation. In general, cyclostationary de- tection can provide a more accurate sensing result and it is robust to variation in the noise power. However, these advantages come at the expense of a higher complexity for implementation and the need for long observation times. Different aspects of cyclostationary detectors are found in [19, 54].

In addition to these methods, recent work has also investigated the use of advanced detection techniques, such as wavelet detectors [19], for per- forming spectrum sensing. Moreover, one can also integrate, in a single secondary system, different detection methods. For example, energy de- tection can be used to perform a fast scan of a wide range of spectrum bands. Subsequently, the results from energy detection can be used to eliminate the spectrum bands with high energy densities (e.g., due to the transmission of PUs). Then, feature detection can be applied to a few can-

Spectrum Sensing in Cognitive Radio Networks

didate bands with low energy densities to search for a unique feature of signals pertaining to PUs.

For measuring the performance of spectrum sensing, three key metrics are explored: The probability of correct detection, the probability of miss and the probability of false alarm. The probability of correct detection is defined as Pd = Prob{decision = H1|H1}, which is the probability of cor-

rectly detecting the transmission of the PU when this PU is active. Subse- quently, the probability of miss is defined as Pm=Prob{decision = H0|H1}

which is the probability of not detecting the PUs transmission while this PU is active, i.e., Pm = 1− Pd. Finally, the probability of false alarm is de-

fined as Pf =Prob{decision = H1|H0} which is the probability of deciding

that the PU is transmitting while the PU is, in fact, idle.

The expressions for computing the different probabilities depend largely on the detection method being employed as well as on the channel condi- tions between the PUs and the SUs. As an example, when one considers energy detection, the probability of false alarm can be given by

Pf = Γ (m,

λ

2)

Γ (m) , (1.10)

where m is the time bandwidth product for energy detection, λ is the energy detection threshold, Γ (·, ·) is the incomplete gamma function, and Γ (·) is the gamma function. Furthermore, for SUs using energy detectors in a Rayleigh fading environment, the average probability of detection can be given by Pd= e−λ2 m_∑−2 n=0 1 n! ( λ 2 )n + ( 1 + ¯γ ¯ γ )m−1[ e− λ 2(1+¯γ) − e−λ2 m_∑−2 n=0 1 n! ( λ¯γ 2(1 + ¯γ) )n] , (1.11) where ¯γ is the average received SNR of the PU signal.

The performance of spectrum sensing is significantly affected by the degradation of the PU signal due to path loss or shadowing. For example, energy or feature detection might be quite affected by a low received SNR from PU signal, due to fading for example. Added to the issue of low SNR is the hidden terminal problem that arises because of shadowing. SUs may be shadowed away from the PU’s transmitter but there may be primary receivers close to the SUs that are not shadowed from the PU transmitter. Thus, if the SU transmits, it may interfere with the primary receiver’s re- ception. Consequently, advanced methods for improving spectrum sensing

Figure 1.5: An illustration of collaborative spectrum sensing in cognitive networks.

are being sought. In particular, it has been shown that, through coopera- tion among the SUs, i.e., collaborative spectrum sensing (CSS), the effects of this hidden terminal problem can be reduced and the probability of detecting the PU can be improved [55–61].

The main idea of CSS is mainly composed of two steps. In the first step, each SU perform its individual detection for spectrum sensing. Then, the SUs would send their sensing bits to a fusion center which, using adequate decision fusion rules, can combined the bits from the different SUs and make a better decision on the presence or absence of the PU. An illustration of a typical CSS approach is shown in Figure 1.5.

The interest in CSS has grown significantly in the past few years. Ex- isting literature has, in fact, studied thoroughly the performance of CSS in cognitive radio networks. For instance, in [55], the SUs perform CSS by sharing their sensing decisions through a centralized fusion center which combines the SUs’ sensing bits using the OR-rule for data fusion. A sim- ilar approach is used in [56] using different decision-combining methods. In [57], it is shown that, in CSS, soft decisions can have an almost com- parable performance with hard decisions while reducing complexity. The authors in [59] propose an evolutionary game model for CSS in order to

Spectrum Sensing in Cognitive Radio Networks

inspect the strategies of the SUs and their contribution to the sensing pro- cess. The effect of the sensing time on the access performance of the SUs in a cognitive network is analyzed in [62]. For improving the performance of CSS, spatial diversity techniques are presented in [58] as a means for combatting the error probability due to fading on the reporting channel between the SUs and the central fusion center. Other interesting perfor- mance aspects of CSS are studied in [60, 61, 63–65].

Existing literature mainly focused on the performance assessment of CSS in the presence of a centralized fusion center that combines all the SUs bits in the network. In practice, the SUs can be at different locations in the network, and, thus, prefer to form nearby groups for CSS without relying on a centralized entity. Moreover, the SUs can belong to different service providers and need to interact with each other for CSS, instead of relaying their bits to a centralized fusion center (which may not even exist in an ad hoc network of SUs). In addition, a centralized approach leads to a significant overhead and increased complexity, notably in large networks. Further, as the number of collaborating SUs increase, the improvement in the probability of detection is accompanied by an increase in the false alarm probability. As a result, given this probability of detection-false alarm tradeoff each SU may only be willing to share their sensing bits with a selected subset. In summary, there is a need for devising models and algorithms that allow the SUs to autonomously interact for performing collaborative spectrum sensing, in a distributed manner, with no need for centralized fusion centers.

For this purpose, in this dissertation, using a coalition formation game formulation, we study the problem of distributed cooperation among the SUs in a cognitive network that seek to improve their sensing performance through CSS. The main contributions and motivations of this work are summarized in Section 8 and the details are found in Paper C.

4.3 Tradeoff between Spectrum Sensing and Spectrum Access