5.1 Foundation and Principle of Cognitive Radio

For any wireless communication system, the radio spectrum is one of the crucial resource constraints and forms the basis for energy-bandwidth tradeoff in wireless systems. It is also a resource with limited availability and restricted access for any specific application or service. To regulate the use of this limited resource, fixed spectrum access policy (FSA) was adopted by spectrum regulators. Under this policy, a fixed piece of spectrum is assigned for a dedicated purpose which is further distributed to the users for access to wireless services. This assigned piece of spectrum with a limited bandwidth is called licensed spectrum and no other user is allowed to use it even though the user with allocation is not using its part. The first decade of the 21^st century experienced an exponential increase in services and applications based on wireless communication which resulted in the allocated spectrum occupying almost an entire range of available radio spectrum and not enough bands were available for upcoming technologies such as Internet of things (IoT) and low power wide area network (LP-WAN). On the other hand, it was reported that a large portion of allocated spectrum was under-utilized whereas some other portions were heavily congested due to allocation with very narrow band gaps, thus resulting in severe adjacent channel interference and drop in quality of service and user experience. The unutilized portions of spectrum are called ‘spectrum holes’ or ‘white spaces’. It can, therefore, be concluded that the scarcity of spectrum is, primarily, due to the lack of flexibility in spectrum allocation and weak optimization in utilization of allocated spectrum.

To overcome the limitation due to scarcity of spectrum and optimize spectrum utilization, techniques based on dynamic spectrum access (DSA), also known as opportunistic spectrum access (OSA), were developed [1, 2]. In DSA, users are categorized as primary users (PUs) and secondary users (SUs). PUs, also known as legacy owners of spectrum, are those users to whom the spectrum is allocated and they have higher priority of using it. SUs are allowed to use the spectrum allocated to PUs when it is not in use. The dynamic allocation of spectrum bands is shown in Figure 5.1. Provision for sharing of the spectrum between PUs and SUs was also introduced in DSA.

To implement DSA, some kind of intelligence is required to sense the utilization of spectrum and identify the not-in-use band. Incorporation of such cognizance about the available spectrum into radio communication devices leads to the evolution of cognitive radio networks (CRNs) and SUs having radio devices with this capability are called cognitive radios (CR). Several surveys and studies on DSA and CR have been conducted in the first decade of the 21^st century [3–5]. In general, CR is a technology which allows for dynamic spectrum utilization based on the need and availability of radio resources, thus optimizing its use with enhanced efficiency [6]. CRs consist of analog RF front end and digital processors which may be dedicated digital signal processors or field programmable gate arrays (FPGAs) customized to function as digital processors. General purpose processors can also be programmed for digital processing of signals received by the front-end. These digital processors are responsible for spectrum sensing. When the spectrum hole available for use by an SU is identified, the software tools running in these digital processors perform all the required changes in parameters of SU such as center frequency and bandwidth of channel to be used, transmit power, wave shape etc. For spectrum sharing between PUs and SUs, three spectrum allocation models are commonly followed—interweave, underlay and overlay [7]. Figure 5.1 shows the interweave model of DSA. In case of underlay model, SUs share the spectrum simultaneously with PUs but at a much lower power level so that there is no or little interference and desired signal can be filtered at the corresponding receiver. In case of overlay, SUs transmit at a much higher power level using the same band even when the PU is active.

Despite the improved spectrum utilization, several limitations and challenges are faced by DSA models which limit their use and deployment for commercial applications. For example, in case of interweave DSA, the SU cannot access a particular band until an active PU exists in that band. This detection of active PU needs to be very accurate to avoid any chances of interference or denial of service. Another challenge is to decide the rate of sensing the spectrum i.e. how recent the sensed data about availability of a spectrum is. Coordination among various PUs and SUs and exchange of control information regarding allocation of spectrum further increase the complexity of the system. Threats by malicious PUs include blocking spectrum bands of inactive PUs, thus causing denial or service to SUs. Similarly, malicious SUs may occupy the available bands and not release it, thus causing severe interference or denial of service for legacy owners or PUs. An SU must release the spectrum band immediately when a signal for its legacy owner PU appears, even though there is an ongoing transmission. This reduces the stability and reliability of the DSA based systems. The techniques to shift to another available band without disrupting the ongoing transmission must be developed to improve these quality of service parameters. In addition to these and many other similar technical challenges, there are several other policies based challenges due to the required coordination and cooperation required between various stakeholders of the overall system. To overcome all these challenges and deployment issues, several techniques and modified forms of DSA are being developed and tested by researchers and industries. These, primarily, include efficient and reliable techniques to sense the spectrum availability in real time and methods to allocate spectrum bands between PUs and SUs maintaining seamless transmission of data and minimizing delay or denial of service. Further sections of this chapter explore such spectrum sensing and allocation techniques.

5.2 Spectrum Sensing for Cognitive Radio Networks

As introduced in section 5.1, several access models have been proposed for the CRs to access radio spectrum using cognitive intelligence. Through these models, spectrum holes are identified by the SUs and used for transmission. For some networks, the SUs may get this information about spectrum holes through broadcasting due to their predetermined availability. However, in most cases, the CR user needs to frequently observe the spectrum to identify the spectrum holes to acquire the spectrum when needed and also to release the in-use spectrum whenever it is demanded by the PU to which it is allocated originally. Such mechanism is known as spectrum sensing [8–10]. To increase the chances of getting access to the vacant spectrum, the CR needs the capability to sense a much wider band of spectrum that it actually requires for communication. This can be implemented by using either an ultra-wideband frontend or multiple frontends sensing different smaller bands. Both these cases will result in an increase in processing time, complexity and hardware cost. Several sensing techniques have been proposed and tested to minimize the effect of these limiting factors with increased efficiency in terms of spectrum sensing capability. The process of spectrum sensing can be described mathematically as shown in equation (5.1).

(5.1)

where S(n) is the signal received at the SU which is a CR user, Gc is the gain of sensing channel, P(n) is the signal from PU and W(n) is the white additive Gaussian noise (AWGN). The two hypotheses corresponding to detection and absence of PU signal are represented by H and H₀ respectively. The signal detected by the frontend of a CR node¹ i.e. S(n) is compared to a threshold (δ) and decision is made using the rule given in equation (5.2).

(5.2)

Based on the hypothesis result, the SU can utilize a particular band at the instant of time if no PU is detected in that band i.e. in the case of H₀. This process of spectrum sensing is shown in Figure 5.2. Several techniques of spectrum sensing have been developed to detect the presence of a PU and vacant spectrum holes.

5.3 Classification of Spectrum Sensing Techniques

Depending upon the source of which is being tested to detect the presence of PU, sensing techniques can be broadly categorized into direct sensing and indirect sensing [8]. Figure 5.3 shows various sensing techniques grouped into direct and indirect sensing.

In case of direct sensing, the CT transmitter searches for the presence of an active primary receiver within its coverage area. Such detection will ensure that the particular frequency band is in use by a PU and thus cannot be used by SU. Techniques such as local oscillator detection, proactive sensing, and interference temperature management have been used for direct sensing. These are also called interference based sensing techniques. The major challenge faced by direct sensing is that the receivers, generally, do not share signals with other users which can be used to identify their active state and spectrum in use. Thus, the task of detecting an active primary transmitter is the preferred approach for spectrum sensing and termed as indirect sensing. In this case, the CR transmitter detects for the presence of an active PU transmitter at a distance which is the sum of its own coverage radius and that of the PU transmitter. Such detection will ensure that the CR transmitter can use the band of detected PU without any interference since chances of active primary receiver within its coverage area will be minimal.

Indirect sensing is further classified into two categories—cooperative sensing and non-cooperative sensing. In non-cooperative sensing, there is no collaboration or co-operation between different SUs and each one takes its own decision regarding presence of PUs. The commonly used non-cooperative sensing techniques are based on various detection mechanisms such as energy detection, matched filter detection, cyclo-stationary detection, Euclidean distance based detection etc. These techniques are easy to implement and benefitted by low processing time and hardware cost. However, these benefits are available at the cost of increase in error rate due to high noise uncertainty, interference and fading. The implementation of these techniques has been briefly discussed in the chapter along with the related research issues.

In case of cooperative sensing, SUs coordinate among themselves while making decision regarding the presence of a PU and availability of spectrum hole [11]. Cooperative sensing is categorized into three types: distributed, centralized and relay assisted. Though drawbacks of non-cooperative sensing are removed using these techniques, they have their own limitations. The succeeding section in this chapter explores non-cooperative sensing techniques.

5.4 Energy Detection

Energy detection is the simplest among all non-cooperative sensing techniques. It does not require any a priori information about the signal of PUs. The sensed energy is used to detect the presence of a PU by comparing it with a pre-decided threshold value which depends upon the noise level. If the received signal is higher than the threshold, the presence of PU signal is concluded. Figure 5.4 shows a basic block diagram of energy detection model. It being a narrow band sensing technique, band pass filter is used to band limit the received signal before it is analyzed for presence of PU signal so that a single frequency channel is tested at a time. The corresponding hypothesis is represented as H₁. The absence of PU signal is represented by hypothesis H₀ [12]. These hypotheses are given in equation (5.3).

(5.3)

where y_SU (n) is the received signal at SU at the output of detector, x_SU(n) is the PU transmitted signal, η(n) is white Gaussian noise and λ is the threshold value.

Before hypothesis testing, the analog signal received (Y_SU(n)) at SU is converted to digital form using digital to analog (D/A) converter and then its fast Fourier transform (FFT) is obtained. The magnitude of FFT is squared and averaged over N samples to obtain y_SU(n) which can be written as shown in equation (5.4). This sampled average is compared with threshold for hypothesis testing [13].

(5.4)

The received signal Y_SU (n) is approximated as a Gaussian random signal and test signal y_SU (n) has central chi-square distribution for H₀ hypothesis and non-central chi-square distribution for H₁ with N degrees of freedom for N > 250 based on the central limit theorem. The Gaussian approximation of SU test signal can be written as shown in equation (5.5).

(5.5)

where ℕ represents normal distribution function, is noise variance and is variance of PU signal. For a channel with additive white Gaussian noise (AWGN), the accuracy of energy detection based sensing technique is measured in terms of its accuracy to detect presence of PU and probability of generating a false alarm regarding presence of a PU signal which may force an SU to release the spectrum resource even though it is available. The probabilities of detection and false alarm can be obtained as given in equations (5.6) and (5.7).

(5.6)

(5.7)

where Q(.) represents Q-function and λ is sensing threshold.

Although energy detection scheme is easy to implement due to it being independent of any prior knowledge of PU signal, its inability to differentiate between signal and noise for low power signals restricts its use in locations which experience strong signal losses or come under deep faded region. The threshold value used for decision making is critical performance parameter for energy detection based sensing schemes. Several schemes have been proposed to dynamically select the threshold in real time and also the decision algorithms [14–18]. These include techniques such as reduction of error in noise power estimation by using dynamic noise estimation [19]. With SINR based linear adaptation for threshold control [20, 21] improvement in SU throughput was achieved but at the cost of increased chances of false alarm. In another adaptive threshold technique based on noise power estimation, the false alarm rate was kept within acceptable limits by using a dedicated noise estimation channel in [22]. Techniques based on wideband spectrum sensing [23], double-threshold [24], filter bank using discrete Fourier transform (DFT) [25] and image binarization [26] have also been used with the objective to reduce false alarm and increase accuracy and probability of detection of PU signal.

5.5 Matched Filter Detection

A linear coherent filter that maximizes the output signal-to-noise ratio (SNR) in the presence of additive noise for a given input signal can be regarded as a matched filter. Let the impulse response of the matched filter be h(n – k) which is the folded time-shifted version of the primary user signal. For the received signal x(k), the basic operation of the matched filter is given by

(5.8)

Matched filter detection is applicable for spectrum sensing in cognitive radio networks only when the a priori knowledge about the center frequency, the bandwidth, the modulation scheme, and other related parameters of the primary user signal as well as the response of the wireless channel are well known. This technique involves the comparison of the received signal with pilot sample signals (or, synchronization codes in some communication systems) received from the same radio transmitter. The test statistic for the matched filter is computed using these saved pilot signals [27, 28].

Let y(n) represent the vector of received signals y(t), and let x_p(n) represent the pilot samples. For N number of samples, the test statistic for matched filter detection technique is given by:

(5.9)

The computed test statistic is then compared with a pre-defined threshold level, δ_MFD. If the computed test statistic is higher than this threshold level, then hypothesis H₁ is valid (i.e., the primary user is present). If the computed test statistic is lower than this threshold level, then hypothesis H₀ is valid (i.e., the primary user is not present). Mathematically, this can be written as:

(5.10)

(5.11)

The value of the pre-defined threshold level, δ_MFD is usually chosen depending on the noise signal level within the incoming received signal. As noise level is quite uncertain, this may result into inaccurate results [29, 30]. Figure 5.5 illustrates a typical block schematic of matched filter detection technique for spectrum sensing [31, 32].

Let us now consider the main performance metrics of spectrum sensing techniques. These are probability of detection (P_d), probability of false detection (P_fd), and probability of miss detection (P_md). The sum of all these performance metrics is unity which determines the overall efficiency of the spectrum sensing techniques. That is,

(5.12)

The probability of detection, P_d signifies that the secondary consumer declares the occurrence of the primary consumer signal when the spectrum is actually busy or occupied. In terms of hypothesis H₁ and H₀ (as defined earlier), these metrics can be stated as:

(5.13)

Higher value of P_d will ensure that there will not be any interference to the primary user signal by the secondary user.

The probability of false alarm detection, P_fd signifies that the secondary user declares the presence of the primary user signal as soon as the spectrum is actually idle or free. In terms of hypothesis H₁ and H₀, these metrics can be expressed as:

(5.14)

Lower value of P_fd will ensure that the secondary users are more likely to access the spectrum, and its higher value may result into interference to the primary user.

The probability of miss detection, P_md signifies that the secondary user declares the non-presence (absence) of the primary user signal when the spectrum is actually busy or occupied. In terms of hypothesis H₁ and H₀, these metrics can be expressed as:

(5.15)

It may be noted that there is always a trade-off between P_fd and P_md, that is, if one is higher, then the other will be lower, or vice versa. A good spectrum sensing technique has to consider the constraints posed by both these metrics. If E_PU represents the primary user signal energy, then the probability of detection, P_d in terms of primary user signal energy, the sensing threshold level and the noise is given by:

(5.16)

where Q(-) is the Q-function, δ_MFD represents the sensing threshold level, and signifies the noise variance. Similarly, the probability of false alarm detection, P_fd is given by:

(5.17)

Alternatively, the sensing threshold level, δ_MFD can be expressed as a function of the primary user signal energy and the noise variance, that is,

(5.18)

In order to enhance the performance of spectrum sensing using matched filter detection technique, dynamic selection of the threshold can be exploited [33]. Although this technique requires a few number of pilot signal samples to obtain good detection performance, it may be impractical sometimes due to non-availability of prior information about the primary user signal. In addition, when the response of the wireless channel changes rapidly (i.e., fading), its performance degrades significantly. This technique may also become unreliable due to presence of primary user emulation security concerns in which a malicious sensor node mimics the primary user signal.

The computational complexity of the matched filter detection is quite high because perfect timing synchronization is needed at both layers— physical as well as medium access control. The detection accuracy can be affected adversely for sensing the spectrum hole in the presence of several primary user signals over the same bandwidth. However, a dedicated matched filter can be used for each primary user signal which, in turn, will increase the complexity of implementing this technique for opportunistic and dynamic spectrum sensing [34–36].

Matched filter detection technique exhibits a good alternative for applications where the transmitted signal is well-known a priori like radar signal processing. To overcome the frequency offset sensitivity, a combination of parallel and segmented matched filter can be used [37]. This arrangement enables the required criteria for sensing time with moderate hardware complexity.

5.6 Cyclo-Stationary Detection

If autocorrelation function of a signal is periodic in nature, then it is termed as a cyclo-stationary signal. The autocorrelation function depends on frequency only in Fourier series expansion. In cyclo-stationary process, different spectral components are not correlated to each other. Almost all communication signals exhibit a statistical property, known as cyclo-stationarity. This can be exploited to determine desired signals and to discriminate against undesired signals, noise signals, and interference in various signal processing techniques without requiring much prior knowledge.

The cyclo-stationary detection technique is primarily based upon the analysis of features of the cyclic autocorrelation function of received periodic data signal and aperiodic noise signals which are regarded as wide-sense stationary signals without any correlation [38]. It is observed that a modulated signal is generally cyclo-stationary, which distinguishes it from noise signals. Some of the statistics such as the mean and autocorrelation function of the modulated signals are periodic in nature [39]. The detection process essentially involves a test on the presence of the cyclo-stationary characteristics of the received primary user signal.

The primary user signal has the periodicity properties which are embedded in its cyclic prefixes, the carrier frequency, modulation rate, pulse trains, hopping sequence, or spreading codes [40]. These are considered to possess cyclo-stationary features whereas the noise is stationary without any correlation. Therefore, this technique is capable of discriminating various primary user signals, the secondary user signals, or the interfering signals (assumed to be either stationary, or, cyclo-stationary with different time periods). Thus, the cyclo-stationary properties are extracted by using either the spectral correlation between the input and output signals, or the cyclic spectrum [41]. The detection of cyclo-stationary signals over multiple cyclic frequencies can be made possible by utilizing generalized likelihood ratio test [42].

The mean, m_y(t) of a cyclo-stationary received signal y(t) is periodic with T₀ as the period of the signal. Mathematically,

(5.19)

Similarly, the autocorrelation function, R_y(t, T₀) of a cyclo-stationary received signal y(t) is periodic with T₀ as the period of the signal. This is expressed mathematically as:

(5.20)

By replacing t with (t + τ/2), and u with (t – τ/2), we write autocorrelation function in terms of cyclic autocorrelation function (CAF), which represents the Fourier series expansion of a cyclo-stationary signal. That is,

(5.21)

Here α represents the cyclic frequency, which is known to the receiver. The value α = 0 results in the standard autocorrelation function of the signal y(t). In fact, a nonzero cyclic frequency (i.e., α ≠ 0) exists for a cyclo-stationary signal such that

Or, the Fourier coefficient,

(5.22)

The cyclic spectral density (CSD), signifies Fourier transform of CAF. It is given by:

(5.23)

The CSD signifies the correlation density between two spectral components. The difference between them is the cyclic frequency. When cyclic spectral density for primary user signals having cyclo-stationary features is computed, the periodicity of their statistics is highlighted. This is quite useful to discriminate the noise signal from the primary user signal. Fast Fourier transform of the correlated signals helps in determining the peak frequencies, indicating the primary user signals even in the presence of noise signals which does not exhibit any periodicities [43, 44]. This means that the CSD has peaks at cyclic frequencies that are multiples of 1/T₀, which happens to be the fundamental frequency of the received primary user signal.

In cyclo-stationary detection, the frequency-domain test statistic for at a specified value of α is given by:

(5.24)

Here the term represents an estimate of the spectral correlation function, and f_s denotes the sampling frequency of the received primary user signal.

In cyclo-stationary detection technique, an analog-to-digital converter is used to convert the received analog signal into digital signal. Fast Fourier transform is then computed using N-point FFT. These values are auto correlated, averaged over number of samples, and then subjected to feature detection. The output of feature detection determines the presence of the primary user (that is, the hypothesis H₁) in case the received signal y(t) is cyclo-stationary (that means the spectrum is occupied), or the absence of the primary user (that is, the hypothesis H₀) in case the received signal y(t) is stationary (that means the spectrum is free). Figure 5.6 illustrates a typical block schematic of cyclo-stationary detection technique.

In cyclo-stationary detection technique, it is assumed that the period T₀ of the received primary user signal is known a priori. If this assumption is not true, then this technique requires either the extraction of, or the search for, the cyclic frequencies. However, this increases the computational complexity. In such case, the Fourier coefficient can be estimated as:

(5.25)

Here signifies the estimation error. It diminishes asymptotically in case of consistent estimator as the observation length tends to approach infinity.

Since the observation length is finite practically, i.e., then the following hypothesis conditions are checked:

• If ; then hypothesis H₀
• If ; then hypothesis H₁

This is possible if autocorrelation functions are transformed into N-vectors by choosing a set of {τ₁,τ₂,………..,τ_N}. The performance of cyclo-stationary detector is analyzed asymptotically [45].

Cyclo-stationary detection is quite robust against noise level uncertainties. It works well even in very low signal-to-noise ratio regions. Due to this, this technique has much less probability of false detection. By increasing the count of samples, the performance of detection can be further enhanced. But this also increases the length of the received signal which, in turn, results in higher sensing time and complexity of implementation. Moreover, sample timing errors and frequency offset may weaken the spectral correlation density [46, 47]. This technique is not robust against fading statistics uncertainties. In a nut-shell, there is a trade-off among performance, complexity, and practicability.

The orthogonal frequency-division multiplexing (OFDM) signals are a typical example of cyclo-stationary signals. Modern wireless communication systems including LTE, WiMAX, Wi-Fi, and DVB-T use spectrum sensing technique with OFDM, and hence use spectrum sensors exploiting cyclo-stationary feature detection [48].

5.9.1 Spectrum and Network Heterogeneity

The spectrum used by CR nodes may spread over widely separated bands which differ in channel characteristics such as coverage area, error rate, path loss, end-to-end delay. This variation accounts for the spectrum heterogeneity and affects the performance of spectrum sharing schemes. Such heterogeneity was introduced in [97] static SUs and fixed channel allocation. Similar approach for both static and mobile users was suggested in [98]. SUs under CRN are being developed with ability to connect using both licensed spectrum bands owned by PUs and unlicensed bands available through various wideband access technologies. Multiple network access modes by different users of a CRN lead to network heterogeneity [99]. Three different access modes are commonly used by CR nodes:

1. Access to primary network: CR users can access the primary cellular network under licensed bands through their access points of base stations. To enable network heterogeneity, MAC protocols are required which can support mobility with horizontal or vertical handover for point of access.
2. Access to CRN: CR users can connect to the CRN using its own CR base station or access points operating on both licensed and unlicensed bands. The spectrum sharing policy in case of communication within the CR need not depend upon primary access techniques.
3. Ad hoc access to CRN: Peer-to-peer or multi-hop communication between various CR nodes can be done in ad hoc mode using licensed or unlicensed spectrum.

This heterogeneity in terms of spectrum and network access raises several challenges and issues in spectrum management for CRN including spectrum sensing, sharing, decision and optimum utilization. Some of these are:

• Common control channel: What radio control parameters must be uncovered and what are the base common interfaces expected to permit interoperability in a heterogeneous situation.
• Control necessities as far as inertness and data move between collaborating hubs.
• • Dynamic radio range: Particular of range server information base and convention interfaces and assessment of execution in dense radio situations.
• Spectrum unit: This is commonly a two-advance procedure: radios need to figure out what recurrence groups are accessible, given the proper FCC rules, and the radios at that point need to choose what band is the most appropriate.
• Location information: Specifically, the wireless devices will need to agree on how to realize various physical, link, and network layer functions in a way that makes best use of the available spectrum, while also satisfying the policy constraints that apply in the selected band.

Issues related to various network operations and actions of CR cycle are discussed further in this section.

5.9.2 Issues and Challenges

The coexistence of primary and CR networks and diverse nature of spectrum utilization by PUs increases the challenges faced by spectrum sensing techniques used in SUs. The CR nodes must adapt to variations in presence of spectrum holes across a wide band of frequencies and remain aware about the spectrum usage by PUs. The spectrum sensing in CR nodes is, generally implemented by exploiting one or more of the three characteristics of spectrum - time, frequency and space. Several aspects of CRN such as intelligence in CR nodes for sensing and decision, overheads in MAC protocols, cross layer implementation of spectrum decision and mobility schemes and flexibility required in CR receiver design contribute to their challenges in design and implementation. Some of these design challenges faced while developing spectrum management algorithms for CRNs are as follows:

• Avoid interference with primary network and its active users.
• Communication with desired QoS levels through dynamic channel with random characteristics and spectrum and network heterogeneity.
• Seamless spectrum and network handoff irrespective of availability or shift of spectrum holes and presence of primary users.
• Efficient MAC protocol for optimization of transmit power and frequency band allocation adaptable to dynamic network conditions.
• • Interference due to coexisting primary and secondary CR users is a challenge to opportunistic communication required for sharing of spectrum.
• Hidden nodes continue to exist and remain undetected by CR nodes due to their low signal strength within the coverage area of SU. This results in additional noise at the SU front end receiver.
• Prediction of spectrum usage by PUs is required for efficient allocation of radio resources to SUs which requires minimum shift from one band to another during an ongoing communication, thus making the link more reliable.
• The complexity involved in implementing spectrum sensing, deciding and allocating schemes for CR needs to be justified in terms of its outcome and benefits achieved. The adoption of CR schemes within the existing regulatory constraints including application domains which restrict the dynamic spectrum allocation is a challenge to be resolved through strategies which are a composite of CR and primary radio schemes [95–100].

The existence of CRNs depends on the ability of CR nodes to decide sense and decide the available spectrum spaces. The placement of such channel sensing, selection/decision and other related performance optimization algorithms is a challenge for researchers. Centralized and distributed are two modes of operation for CR systems. Several sensing and allocation/decision algorithms have been discussed in this chapter. However, their customized or modified forms are being developed to resolve the issues related to performance improvement and implementation in unexplored application areas. The extensive growth of machine learning has helped the CR nodes in various aspects of cognitive cycle. The choice of supervised or unsupervised learning is the first challenge faced while making a suitable decision for adopting machine learning algorithms. The design and implementation of such algorithms may further increase the complexity of CR systems. Another challenge is associated with implementing geo-location technology which is by CR systems to find and track the location of mobile PUs and SUs, thus contributing to estimation of resource requirement and availability of spectrum holes or chances of a PU entering the coverage area of an SU. The challenges are related to implementation and use of database along with acquiring accurate location of CR nodes. It becomes more challenging in case of the nodes which are not equipped with GPS.

Security of devices and users (identity and data) are the foremost requirements for any network based system. This also includes compliance of regulations. The related challenges include device access authorization, attempts of violation and potential threats, security certifications of software resources and presence of malicious software in the device. Front-end of CR nodes is another major source of challenges faced by CR designers. This includes frequency agile transceivers with wide-band operation, low noise amplifier (LNA) with high linearity in wide band and impedance matching for a variety of device outputs. The ADC and DAC are integral parts of CR systems since processing within the CR devices is conducted in digital domain whereas the RF signal at the front-end is analog. The need to operate these ADC and DAC over a wide spectrum and at several power levels is a challenge for researchers. Other related challenges are calibration of channel mismatch for difference spectrum bands occupied by CR nodes at different times and skew error correction in sensing and front-end tuning [101]. Techniques such as post-linearization in digital domain and interleaving time and frequency multiplexing using set-pass sampling filter are being used to reduce complexity, increase sensitivity with less imperfections. The design issues at base-band level architecture include flexible operation with optimized tradeoff between power consumption and performance. This demands for reconfigurable resources, efficient power management schemes and dynamic task management.

The issues and challenges highlighted are crucial for effective deployment of CRNs for real-time communication systems. The challenges are multiple depend upon the usage scenarios and bounds of physical systems. Further research is required to develop algorithms, protocols and schemes which can resolve the issues and challenges faced by the current implementation of CRN and future technologies.

1. 1. Zhang, Y., Dynamic Spectrum Access in Cognitive Radio Wireless Networks. 2008 IEEE International Conference on Communications, Beijing, pp. 4927– 4932, 2008.
2. 2. Song, M., Chunsheng, X., Yanxiao, Z., Xiuzhen, C., Dynamic spectrum access: From cognitive radio to network radio. IEEE Wirel. Commun., 19, 23–29, 2012.
3. 3. Akyildiz, I.F., Lee, W.Y., Vuran, M.C., Mohantly, S., Next generation/dynamic spectrum access/cognitive radio wireless network: A survey. Elsevier Comput. Netw., 50, 2127–2159, 2006.
4. 4. Raychaudhuri, D. et al., Cognet: An Architectural Foundation For Experimental Cognitive Radio Networks Within the Future Internet. Proc. Mobi. Arch., 11–16, 2006.
5. 5. Cormio, C. and Chowdhury, K.R., A survey on MAC protocols for cognitive radio networks. Ad Hoc Netw., 7, 7, 1315–1329, 2009.
6. 6. Mitola, J. and Maguire, G.Q., Cognitive radio: Making software radios more personal. IEEE Pers. Commun., 6, 4, 13–18, 1999.
7. 7. Goldsmith, et al., Breaking Spectrum Gridlock with Cognitive Radios: An Information Theoretic Perspective. Proc. IEEE, 97, 5, 894–914, 2009.
8. 8. Kaabouch, N. and Hu, W.-C., Handbook of Research on Software-Defined and Cognitive Radio Technologies for Dynamic Spectrum Management. IGI Glob. J., 2014.
9. 9. Salahdine, F., Kaabouch, N., El Ghazi, H., A Real Time Spectrum Scanning Technique based on Compressive Sensing for Cognitive Radio Networks. The 8th IEEE Annual Ubiquitous Computing, Electronics & Mobile Commun. Conf., pp. 1–6, 2017.
10. 10. Yucek, T. and Arslam, H., A Survey of Spectrum Sensing Algorithms for Cognitive Radio Applications. IEEE Commun. Surv. Tutorials 11, 1, 116–130, 2009.
11. 11. Ben Letaief, K. and Zhang, W., Cooperative Spectrum Sensing. Cogn. Wirel. Commun. Netw., 115–138, 2007.
12. 12. Cabric, D., Mishra, S.M., Brodersen, R.W., Implementation issues in spectrum sening for cognitive radios. Proceedings of the Thirty-Eighth Asilomar Conference on Signals, vol. 1, pp. 772–776, 2004.
13. 13. Manesh, M.R., Apu, S., Kaabouch, N., Hu, W., Performance Evaluation of Spectrum Sensing Techniques for Cognitive Radio Systems. Ubiquit. Comput. Electron. Mobile Commun. Conf. IEEE Annu., 19, 1–6, 2016.
14. 14. Rawat, P., Singh, K.D., Bonnin, J.M., Cognitive radio forM2Mand Internet of Things: A survey. Comput. Commun., 94, 1–29, 2016.
15. 15. Plata, D.M.M. and Reátiga, Á.G.A., Evaluation of energy detection for spectrum sensing based on the dynamic selection of detection-threshold. J. Procedia Eng., 35, 135–143, 2012.
16. 16. Joshi, D.R., Popescu, D.C., Dobre, O.A., Adaptive spectrum sensing with noise variance estimation for dynamic cognitive radio systems. Proceedings of the Conference on Information Sciences and Systems, Princeton, NJ, USA, 17–19 March, pp. 1–5, 2010.
17. 17. Muralidharan, A., Venkateswaran, P., Ajay, S.G., Prakash, D.A., Arora, M., Kirthiga, S., An adaptive threshold method for energy-based spectrum sensing in Cognitive Radio Networks. Proceedings of the International Conference on Control, Instrumentation, Communication, and Computational Technologies, Kumaracoil, India, 18–19 December, pp. 8–11, 2015.
18. 18. Sarker, M., Energy detector based spectrum sensing by adaptive threshold for low SNR in CR networks. Proceedings of the Wireless and Optical Communication Conference, Taipei, Taiwan, 23–24 October, pp. 118–122, 2015.
19. 19. Khan, A.A., Rehmani, M.H., Reisslein, M., Cognitive radio for smart grids: Survey of architectures, spectrum sensing mechanisms, and networking protocols. IEEE Commun. Surv. Tutorials, 18, 1, 860–898, 2016.
20. 20. Vartiainen, J., Sarvanko, H., Lehtomaki, J., Juntti, M., Latva-Aho, M., Spectrum Sensing with LAD-Based Methods. IEEE 18th Int. Sym., 1–5, 2007.
21. 21. Manesh, M.R., Kaabouch, N., Reyes, H., A Bayesian Approach to Estimate and Model SINR in Wireless Networks. Int. J. Commun. Syst. Wiley, 1–11, 2016.
22. 22. Gong, S., Wang, P., Liu, W., Spectrum sensing under distribution uncertainty in cognitive radio networks. IEEE Int. Conf. Commun., 1512–1516, 2012.
23. 23. Lim, C.H., Adaptive energy detection for spectrum sensing in unknown white Gaussian noise. IET Commun., 6, 1884–1889, 2012.
24. 24. Xiang, L., Bin, W., Hong, W., Pin-Han, H., Zhiqiang, B., Lili, P., Adaptive Threshold Control for Energy Detection Based Spectrum Sensing in Cognitive Radios. IEEE Wirel. Commun. Lett., 1, 448–451, 2012.
25. 25. Joshi, D.R., Popescu, D.C., Dobre, O.A., Adaptive spectrum sensing with noise variance estimation for dynamic cognitive radio systems. Proceedings of the Conference on Information Sciences and Systems, Princeton, NJ, USA, 17–19 March, 2010.
26. 26. Muralidharan, A., Venkateswaran, P., Ajay, S.G., Prakash, D.A., Arora, M., Kirthiga, S., An adaptive threshold method for energy-based spectrum sensing in Cognitive Radio Networks. Proceedings of the International Conference on Control, Instrumentation, Communication, and Computational Technologies, Kumaracoil, India, 18–19 December, pp. 8–11, 2015.
27. 27. Salahdine, F., El Ghazi, H., Kaabouch, N., Fihri, W.F., Matched filter detection with dynamic threshold for cognitive radio networks. 2015 International Conference on Wireless Networks and Mobile Communications (WINCOM), IEEE, pp. 1–6, 2015.
28. 28. Zhang, X., Chai, R., Gao, F., Matched filter based spectrum sensing and power level detection for cognitive radio network. 2014 IEEE global conference on signal and information processing (GlobalSIP), IEEE, pp. 1267–1270, 2014.
29. 29. Jiang, C., Li, Y., Bai, W., Yang, Y., Hu, J., Statistical matched filter based robust spectrum sensing in noise uncertainty environment. 2012 IEEE 14th International Conference on Communication Technology, IEEE, pp. 1209– 1213, 2012.
30. 30. Lv, Q. and Gao, F., Matched filter based spectrum sensing and power level recognition with multiple antennas. 2015 IEEE China Summit and International Conference on Signal and Information Processing (ChinaSIP), IEEE, pp. 305– 309, 2015.
31. 31. Manesh, M.R., Subramaniam, S., Reyes, H., Kaabouch, N., Real-time spectrum occupancy monitoring using a probabilistic model. Comput. Netw., 124, 87–96, 2017.
32. 32. Avinash, P., Gandhiraj, R., Soman, K.P., Spectrum sensing using compressed sensing techniques for sparse multiband signals. Int. J. Sci. Eng. Res., 3, 5, 2012.
33. 33. Sharma, S.K., Bogale, T.E., Chatzinotas, S., Le, L.B., Wang, X., Ottersten, B., Improving robustness of cyclostationary detectors to cyclic frequency mismatch using Slepian basis. 2015 IEEE 26th Annual International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC), IEEE, pp. 456–460, 2015.
34. 34. Cabric, D., Mishra, S.M., Brodersen, R.W., Implementation issues in spectrum sensing for cognitive radios. Conference Record of the Thirty-Eighth Asilomar Conference on Signals, Systems and Computers, vol. 1, pp. 772–776, 2004.
35. 35. Chen, H.-S., Gao, Wen, Daut, D.G., Signature based spectrum sensing algorithms for IEEE 802.22 WRAN. 2007 IEEE International Conference on Communications, IEEE, pp. 6487–6492, 2007.
36. 36. Sahai, A., Hoven, N., Tandra, R., Some fundamental limits on cognitive radio. Allerton Conference on Communication, Control, and Computing, pp. 1662–1671, 2004.
37. 37. Zhang, Z., Yang, Q., Wang, L., Zhou, X., A novel hybrid matched filter structure for IEEE 802.22 standard. 2010 IEEE Asia Pacific Conference on Circuits and Systems, IEEE, pp. 652–655, 2010.
38. 38. Gardner, W.A., Exploitation of spectral redundancy in cyclostationary signals. IEEE Signal Process. Mag., 8, 2, 1991.
39. 39. Gardner, W.A., An introduction to cyclostationary signals, in: Cyclostationarity in communications and signal processing, pp. 1–90, IEEE press, New York, 1994.
40. 40. Akyildiz, I.F., Lo, B.F., Balakrishnan, R., Cooperative spectrum sensing in cognitive radio networks: A survey. Phys. Commun., 4, 1, 2011.
41. 41. Atapattu, S., Tellambura, C., Jiang, H., Energy detection for spectrum sensing in cognitive radio, Springer, New York, NY, USA, 2014.
42. 42. Lundén, J., Koivunen, V., Huttunen, A., Poor, H.V., Collaborative cyclostationary spectrum sensing for cognitive radio systems. IEEE Trans. Signal Process., 57, 11, 2009.
43. 43. Gardner, W.A., An Introduction to Cyclostationary Signals, Chapter 1, Cyclostationarity in Communications and Signal Processing, IEEE Press, Piscataway, NJ, 1993.
44. 44. Cabric, D., Mishra, S.M., Brodersen, R.W., Implementation issues in spectrum sensing for cognitive radios, in: Conference Record of the Thirty-Eight Asilomar Conference on Signals, Systems and Computers, vol. 1, pp. 772–776, 2004.
45. 45. Dandawate, A.V. and Giannakos, G.B., Statistical tests for presence of cyclostationarity. IEEE Trans. Signal Process., 42, 9, 2355–2369, 1994.
46. 46. Fan, R. and Jiang, H., Optimal multi-channel cooperative sensing in cognitive radio networks. IEEE Trans. Wirel. Comm., 9, 3, 1128–1138, 2010.
47. 47. Zeng, Y., Liang, Y.-C., Hoang, A.T., Zhang, R., A review on spectrum sensing for cognitive radio: Challenges and solutions. EURASIP J. Adv. Signal Process., 1, 2010.
48. 48. Goldsmith, A., Wireless communications, Chapter 12, Cambridge University Press, Stanford University, California, 2005.
49. 49. Mitola, J. and Maguire, G.Q., Cognitive radio: Making software radios more personal. IEEE Pers. Commun., 6, 4, 13–18, 1999.
50. 50. Hu, H., Zhang, H., Yu, H., Delay QoS guaranteed cooperative spectrum sening in cognitive radio networks. AEU-Int. J. Electron. Commun., 67, 9, 804–807, 2013.
51. 51. Mukherjee, A., Choudhury, S., Goswami, P., Bayessa, G.A., Tyagi, S.K.S., A novel approach of power allocation for secondary users in cognitive radio networks. Comput. Electr. Eng., 75, 301–308, 2019.
52. 52. Salahdine, F., El Ghazi, H., Kaabouch, N., Fihri, W.F., Matched filter detection with dynamic threshold for cognitive radio networks. 2015 International Conference on Wireless Networks and Mobile Communications (WINCOM), IEEE, pp. 1–6, 2015.
53. 53. Khan, A.A., Rehmani, M.H., Reisslein, M., Cognitive radio for smart grids: Survey of architectures, spectrum sensing mechanisms, and networking protocols. IEEE Commun. Surv. Tutorials, 18, 1, 860–898, 2015.
54. 54. Manesh, M.R., Quadri, A., Subramaniam, S., Kaabouch, N., An optimized SNR estimation technique using particle swarm optimization algorithm. 2017 IEEE 7th Annual Computing and Communication Workshop and Conference (CCWC), IEEE, pp. 1–6, 2017.
55. 55. Quan, Z., Cui, S., Sayed, A.H., Poor, H.V., Wideband spectrum sensing in cognitive radio networks, Proceedings of the 2008 IEEE International Conference on Communications, Beijing, May 19–23, 2008. arXiv preprint arXiv:0802.4130.
56. 56. Subramaniam, S., Reyes, H., Kaabouch, N., Spectrum occupancy measurement: An autocorrelation based scanning technique using USRP. 2015 IEEE 16th Annual Wireless and Microwave Technology Conference (WAMICON), IEEE, pp. 1–5, 2015.
57. 57. Naraghi-Pour, M. and Ikuma, T., Autocorrelation-based spectrum sensing for cognitive radios. IEEE Trans. Veh. Technol., 59, 2, 718–733, 2009.
58. 58. Almasaeid, H.M., Spectrum Allocation Algorithms for Cognitive Radio Mesh Networks, Graduate Theses and Dissertations, p. 10383, 2011.
59. 59. Kasbekar, G.S., Economics of Spectrum Allocation in Cognitive Radio Networks, Publicly Accessible Penn Dissertations, p. 446, 2011.
60. 60. Zhao, Z., Peng, Z., Zheng, S., Shang, J., Cognitive radio spectrum allocation using evolutionary algorithms. IEEE Trans. Wirel. Comm., 8, 9, 4421–4425, 2009.
61. 61. Akyildiz, I.F., Lee, W., Vuran, M.C., Mohanty, S., A survey on spectrum management in cognitive radio networks. IEEE Commun. Mag., 46, 4, 40–48, 2008.
62. 62. Akbar, I.A. and Tranter, W.H., Dynamic spectrum allocation in cognitive radio using hidden Markov models: Poisson distributed case. Proceedings 2007 IEEE South east Con, Richmond, pp. 196–201, 2007.
63. 63. Akhoondi, F., Poursaeed, O., Salehi, J.A., Resource allocation using fragmented-spectrum synchronous OFDM-CDMA in cognitive radio networks. 2014 Iran Workshop on Communication and Information Theory (IWCIT), Tehran, 2014.
64. 64. Kokila, V. and Bharathi, A.J., Fragmented Spectrum Allocation in Cognitive Radio Networks for Energy Efficiency and Improved Throughput. J. Eng. Technol., 9–15, 2015.
65. 65. Koroupi, F., Talebi, S., Salehinejadb, H., Cognitive radio networks spectrum allocation: An ACS perspective. Sci. Iran., 19, 3, 767–773, 2012.
66. 66. Zhao, Z., Peng, Z., Zheng, S., Shang, J., Cognitive radio spectrum allocation using evolutionary algorithms. IEEE Trans. Wirel. Comm., 8, 4421–4425, 2009.
67. 67. Zhou, X., Li, G., Li, D., Wang, D., Soong, K., Probabilistic resource allocation for opportunistic spectrum access’’. IEEE Trans. Wirel. Comm., 99, 1–10, 2010.
68. 68. Wang, J., Huang, Y., Jiang, H., Improved algorithm of spectrum allocation based on graph coloring model in cognitive radio. Int. Conf. Commun. Mobile Comput., 353–357, 2009.
69. 69. Liu, Y., Xu, G., Tan, X., A novel spectrum allocation mechanism based on graph coloring and bidding theory. Int. Conf. Comput. Intell. Nat. Comput., 155–158, 2009.
70. 70. Borkotoky, S.S., Kottapalli, S.L., Pursley, M.B., Analytical Methods for Performance Evaluations of Adaptive Modulation and Coding in Cognitive Radio Systems That Employ Distance Statistics. IEEE Trans. Cogn. Commun. Netw., 5, 1, 2019.
71. 71. Anumandla, K.K., Peesapati, R., Sabat, S.L., Field programmable gate array implementation of spectrum allocation technique for cognitive radio networks. Comput. Electr. Eng., 178–192, 2015.
72. 72. Bin Shahid, M.I., Kamruzzaman, J., Hassan, Md. R., Modeling multiuser spectrum allocation for cognitive radio networks. Comput. Electr. Eng., 2015.
73. 73. Sandeep, A Comparative Analysis of Optimization Techniques in Cognitive Radio (QoS). Int. J. Eng. Adv. Technol., 6, 3, 2017.
74. 74. Steenkiste, P., Sicker, D., Minden, G., Raychaudhuri, D., Future Directions in Cognitive Radio Network Research, 2009, NSF Workshop Report.
75. 75. Kirar, V.P.S., Artificial Neural Networks for Cognitive Radio Network: A Survey. Int. J. Electron. Commun. Eng., 9, 1, 2015.
76. 76. Salau, H.B. et al., An Examination of Application of Artificial Neural Network in Cognitive Radios. IOP Conf. Ser.: Mater. Sci. Eng., 53, 012036, 2013.
77. 77. Jamal, E., Dynamic Spectrum Allocation in Cognitive Radio Cellular Networks. WSEAS Trans. Commun., 16, 2017.
78. 78. Zhou, F., Zhang, X., Hu, R.Q., Papathanassiou, A., Meng, W., Resource Allocation Based on Deep Neural Networks for Cognitive Radio Networks. 2018 IEEE/CIC International Conference on Communications in China (ICCC), Beijing, China, pp. 40–45, 2018.
79. 79. Tuberquia-David, L.M., Cruz, L., Hernández, C., Spectral Prediction: Approaches in Cognitive Radio Networks. Int. J. Appl. Eng. Res., 13, 10, 2018.
80. 80. Tan, X., Huang, H., Ma, L., Frequency allocation with Artificial Neural Networks in cognitive radio system. IEEE 2013 Tencon - Spring, Sydney, NSW, pp. 366–370, 2013.
81. 81. Lee, W., Resource Allocation for Multi-Channel Underlay Cognitive Radio Network Based on Deep Neural Network. IEEE Commun. Lett., 22, 9, 1942– 1945, 2018.
82. 82. Chen, M., Challita, U., Saad, W., Yin, C., Debbah, M., Machine Learning for Wireless Networks with Artificial Intelligence: A Tutorial on Neural Networks, 2017.
83. 83. Anghuwo, A.A., Liu, Y., Tan, X., Liu, S., Spectrum Allocation Based on Game Theory in Cognitive Radio Networks, in: Information and Automation. ISIA 2010. Communications in Computer and Information Science, vol. 86, L. Qi (Ed.), Springer, Berlin, Heidelberg, 2011.
84. 84. Pai, V., Prabhu, A., Menon, A., Game Theoretic optimization of spectrum allocation in cognitive radio. International Conference on Computing Communication Control and automation (ICCUBEA), Pune, pp. 1–5, 2016.
85. 85. Musku, M.R. and Cote, P., Time Domain Spectrum Allocation using Game Theory for Cognitive Radios. Proceedings of the 9th International Conference on Mobile and Wireless Communications Networks, Cork, Ireland, September 19-21, 2007.
86. 86. Ni, Q., Zhu, R., Wu, Z., Sun, Y., Zhou, L., Zhou, B., Spectrum Allocation Based on Game Theory in Cognitive Radio Networks. J. Netw., 8, 3, 2013.
87. 87. Hossain, K., El-Saleh, Md. A., Ayman, Cognitive Radio Engine Model Utilizing Soft Fusion Based Genetic Algorithm For Cooperative Spectrum Optimization. Int. J. Comput. Netw. Commun., 2013.
88. 88. Chatterjee, S., Dutta, S., Bhattacharya, P., Roy, J., Optimization of Spectrum Sensing Parameters in Cognitive Radio Using Adaptive Genetic Algorithm. J. Telecommun. Inf. Technol., 1, 21–27, 2017.
89. 89. Elhachmi, J. and Guennoun, Z., Cognitive radio spectrum allocation using genetic algorithm. Elhachmi and Guennoun EURASIP J. Wirel. Commun. Netw., 2016.
90. 90. Hou, B., Zu, Y., Li, W., Liu, G., Ding, J., Cognitive Radio Spectrum Allocation Strategy Based on Improved Genetic Algorithm. Commun. Netw., 2013.
91. 91. Devi, M., Sarma, N., Deka, S.K., Spectrum Allocation in Cognitive Radio Networks—A Centralized Approach, in: Advanced Computational and Communication Paradigms. Lecture Notes in Electrical Engineering, vol. 475, Springer, Singapore, 2018.
92. 92. Singh, W.J. and Marchang, N., A Review on Spectrum Allocation in Cognitive Radio Network. Int. J. Commun. Netw. Distr. Syst., 2018.
93. 93. Thomas, A.T., Vergados, D.D., Siakavara, K., Nikolaidis, S., Goudos, S.K., Sarigiannidis, P., Obaidat, M., Spectrum allocation in cognitive radio networks using chaotic biogeography-based optimization. IET Netw., 2018.
94. 94. Rahama, Y.A., Hassan, M.S., Ismail, M.H., A Stochastic-Based Rate Control Approach for Video Streaming Over Cognitive Radio Networks. IEEE Trans. Cogn. Commun. Netw., 5, 1, 2019.
95. 95. Wang, D. and Song, Y., ECCO: A Novel End-to-End Congestion Control Scheme in Multi-Hop Cognitive Radio Ad Hoc Networks. IEEE Trans. Cogn. Commun. Netw., 5, 1, 2019.
96. 96. Ning, G., Cao, X., Duan, J., Chowdhury, K.R., A Spectrum Sharing Algorithm Based on Spectrum Heterogeneity for Centralized Cognitive Radio Networks. 2011 IEEE 73rd Vehicular Technology Conference (VTC Spring), Yokohama, pp. 1–5, 2011.
97. 97. Miao, M. and Tsang, D.H.K., Impact of Channel Heterogeneity on Spectrum Sharing in Cognitive Radio Networks. Proc. IEEE ICC, 2377–2382, 2008.
98. 98. Qiu, R.C., Hu, Z., Li, H., Wicks, M.C., Cognitive Radio Communications and Networks Principles and Practice, John Wiley & Sons Ltd, Wiley Online Library, 2012.
99. 99. Akyildiz, F., Lee, W., Chowdhury, K.R., Spectrum management in cognitive radio ad hoc networks. IEEE Netw., 23, 4, 6–12, 2009.
100. 100. Kamil, A.S. and Khider, I., Open Research issues in Cognitive Radio. 16th Telecommunications Forum TELFOR, Serbia, Belgrade, 25-27 Nov, 2008.
101. 101. Gruget, M.R., Nguyen, V.T., Lelandais-Perrault, C., Bénabès, P., Loumeau, P., Wide-band multipath A to D converter for cognitive radio applications. Proceedings of the IEEE International Microwave Workshop Series on RF Front-Ends for Software Defined and Cognitive Radio Solutions (IMWS ‘10), Aveiro, Portugal, pp. 73–76, 2010.
102. 102 Garg, P.K., Dua, T.R., Chandra, A., Spectrum Challenges for Modern Mobile Services, J. ICT Standardization, 1, 137–158, 2013.