TABLE OF CONTENTS
Decleration of Authorship
List of Figures
List of Tables
2.1 Cognitive Radio Overview
2.2 Spectrum Sensing
2.2.1 Matched Filter
2.2.2 Cyclostationary Feature Detector
2.2.3 Energy Detector
2.2.4 Auto-correlation Based Detector
2.2.5 Cooperative Detector
2.3 Literature Survey
2.3.1 Binary Hypothesis
2.3.2 Neyman-Pearson Test
2.3.3 Sequential Test
2.3.4 Sequential probability Ratio Test (SPRT)
2.3.5 Truncated Sequential probability Ratio Test (TSPRT)
3.1 Introduction to SIMULINK
3.2 Flow chart describing model
3.3 SIMULINK model
4. Xilinx Implementation Implementation
4.1 Brief about FPGA
4.2 Introduction to system generator in XILINX
4.3 Process Flow
4.4 Xilinx Model
5. Results and Analysis
5.1 ADC bit variation and its effect on ASN
5.2 Variation of noise parameter σn
5.3 Effect of q/σ and dynamic range of signal
5.4 Detecting false alarms
7. Limitations and Future Work
To ensure that cognitive radios would not interfere with primary users, spectrum sensing is required to be efficient and accurate by reliably detecting primary user signals. In this work, we implemented a spectrum sensing methodology based on Truncated Sequential Probability Ratio Test (TSPRT). TSPRT is a combination of SPRT and Neyman-Pearson. We created and simulated the model and observed the variation of quantization error, noise variance and dynamic range of the signal to achieve the minimum average sample number (ASN) and desired error probabilities of detection and false alarm for sine wave and similar input signals. This report comprises of theoretical analysis and practical implementation of spectrum sensing circuit in Xilinx system generator. Simulations are done to observe the effect of various parameters on ASN and shown
It is my pleasure to record deepest gratitude to my guide Prof. G.S.Visveswaran for giving me this opportunity to work under his supervision and their interest and valuable suggestions
I would like also to thank members of the cyber lab, Impact Lab, my classmate Divya and all my well wishers for their constant guidance and support throughout this project
LIST OF FIGURES
Figure 1.1 Measurement of 0-6 GHz spectrum utilization at BWRC
Figure 2.1 Conventional Energy Detection with Single Threshold
Figure 3.1 Flowchart for SIMULINK Model
Figure 3.2 SIMULINK model of Cogitive radio Spectrum Sensing Circuit
Figure 4.1 Xilinx System Generation Design Flow
Figure 4.2 Main Model built in Xilinx system Generator Generator
Figure 4.3 Signal Processing Sub block
Figure 4.4 Pf and Pm calculation sub block
Figure 4.5 NT and VT calculation Sub block
Figure 4.6 AT and BT calculation Sub block
Figure 4.7 Final Decision Circuitry
LIST OF TABLES
Table 3.1 Noise variance and noise power
Table 5.1 Variation of ADC bits
Table 5.2 Variation of σn
Table 5.3 Variation of q/σ and dynamic range
Table 5.4 False alarm observation
CHAPTER 1 INTRODUCTION
As the use of radio spectrum is increasing day by day, its scarcity is increasing and becoming a concerned issue. In most of the countries frequency bands below 3GHz range has been allocated to specific uses, which are particularly valuable due to their favorable propagation characteristics for long range communication. While it has been observed that frequency bands in the range 0-2.5 GHz are unevenly utilized ranging from 15% to 85%, this shows very low spectrum utilization efficiency. According to recent reports released by the Federal Communications Commission1 show that, a large amount of allocated spectrum particularly television bands, is substantially under-utilized most of the time whereas a small portion of spectrum bands such as cellular bands, experience increasingly congestion and scarcity due to rapid deployment of various wireless services. There are two frequency bands 400-800 MHz (UHF TV bands) and 3-10 GHz in which cognitive radios might operate2. The FCC has noted that in the lower UHF bands almost every geographical area has several unused 6 MHz wide TV channels. Furthermore, given the static TV channel allocations, the timing requirements for spectrum sensing are very relaxed, this band has very low spectral utilization, as indicated in Figure 1.1. Measurements show that primary users who have been allotted frequency bands by regulatory authority are not using it all the time. Hence other users may use this spectrum for their use but they do not have the license to use it. So for this purpose, Cognitive radio emerged so as to provide opportunistic spectrum sharing.
In opportunistic spectrum sharing secondary users are allowed to operate in specific frequency bands without hampering the primary users consent and therefore can dramatically improve spectrum utilization. Care has to be taken while altering the spectrum as licensed users are prior to unlicensed (secondary) users and hence Quality of Service (QoS) of primary users is required to be maintained. Hence spectrum sensing performed by secondary users (SUs) to detect the unoccupied frequency bands, is the key enabling technique to meet this requirement. Spectrum sensing model needs to be effective and precise that can cope with several critical challenges of cognitive radio networks like , it is often difficult for Secondary users in a cognitive radio network to acquire knowledge about primary users signals. The Detection System should be able to quickly detect primary signals at reasonably low Signal-to-Noise (SNR) ratios and with low false and missed alarm probabilities. There are so many schemes for spectrum sensing, such as matched-filter detection, energy detection and cyclostationary detection3, proposed and investigated. Among these sensing schemes, energy detection is very good as it does not rely on any deterministic knowledge of the primary signals and has low implementation complexity. However, when the detection SNR is low, energy detection entails a large amount of sensing time to ensure high detection accuracy, e.g., the sensing time is inversely proportional to the square of SNR4. To overcome this shortcoming, several sensing schemes based on the sequential probability ratio test (SPRT) have been proposed under various cognitive radio settings.
illustration not visible in this excerpt
Figure 1.1: Measurement of 0-6 GHz spectrum utilization at BWRC
The main advantage of the SPRT is that, for given detection error probabilities, the SPRT requires the smallest average sample number (ASN) for testing simple hypotheses. In comparison to fixed-sample-size sensing schemes, the sensing scheme based on the SPRT requires much reduced sensing time on the average while maintaining the same detection performance5. But SPRT based sensing scheme also suffers from potential drawbacks for e.g. First, Sometimes ASN tends to a very large value which incurs high time complexity, evaluating the probability ratio requires deterministic knowledge or statistical distribution of certain parameters of the primary signals. Acquiring such deterministic information or statistical distribution is practically difficult in general. Thirdly, the existing SPRT based sensing scheme adopts the Wald’s choice on the thresholds [6+. However, the Wald’s choice, which works well for the non- truncated SPRT, increases error probabilities when applied to the truncated SPRT.
In this project, we devised a model based on Truncated Sequential Probability Ratio Test (TSPRT), based on the sequential shifted chi-square test. The TSPRT possesses several attractive features:
- Like energy detection, the TSPRT only requires the knowledge on noise power and does not rely on any deterministic knowledge about primary signals
- TSPRT is capable of delivering considerable reduction on the average sensing time while maintaining a comparable detection performance as compared to fixed-sample-size detection such as energy detection,
- In comparison with the SPRT based sensing scheme, the TSPRT has a much simpler test statistic and thus has lower implementation complexity.
- And TSPRT offers desirable flexibility to strike a trade-off between detection performance and sensing time when the operating SNR is higher than the minimum detection SNR.
So theoretical concepts and formulations are taken from literature survey, and since their results rely on theoretical analysis and computer simulations, in order to approve this technology and fully understand the system design issues, these theoretical results are verified and demonstrated in realistic scenarios through model implementation in MATLAB/SIMULINK using Xilinx block set and various performance parameters are analyzed to get the minimum possible value average sample number to detect the signal and error probabilities.
CHAPTER 2 BACKGROUND
2.1 Cognitive Radio Overview
Cognitive radios have to sense the spectrum to detect the presence or absence of primary user signals. Secondary users are allowed to operate in frequency bands without the consent of the Primary users of these bands. The spectrum has to be sensed accurately to find out even weaker primary user signals. At the same time, cognitive radios have to respect the needs of the primary users (PUs) and not to interfere with them. Therefore, the spectrum sensing method has to be very sensitive and distinguish PU signals below the noise floor. Many different spectrum sensing methods have been introduced 3. One of the simplest spectrum sensing methods is the energy detection. Energy detectors have been introduced nearly half a century ago by Urkowitz7, and they are still researched and new ways to enhance their efficiency are published. Energy detectors do not need any information about the signal under detection; therefore they are able to detect wide variety of signals. However, they cannot differentiate primary user’s signals from noise. Other signal detection method exploits the statistical properties of PU signals to detect them. One of these methods is cyclostationary spectrum sensing. Cyclostationary feature detectors can differentiate noise from primary user’s signals. There are Algorithms which can sense even more specific signals by matched filtering. Matched filters deliver optimal detection performance; however, each signal under detections needs a specific matched filter. For this reason, matched filter is not widely used.
2.2 Spectrum Sensing
Spectrum sensing is requires to find out whether the required frequency band is free or not so as to be utilized by secondary users. The spectrum has to be sensed accurately to find out even weaker primary user signals. At the same time, cognitive radios have to respect the needs of the primary users (PUs) and not to interfere with them. Therefore, the spectrum sensing method has to be very sensitive and distinguish PU signals below the noise floor. Three schemes are generally used for sensingǱ matched filter detection, energy detection and cyclostationary feature detection3.
2.2.1 Matched Filter Detection
When the information of the primary user signal is known to the secondary user, the optimal detector in stationary Gaussian noise is the matched filter since it maximizes the received signal-to-noise ratio (SNR). While the main advantage of the matched filter is that it requires less time to achieve high processing gain due to coherency, it requires a priori knowledge of the primary user signal such as the modulation type and order, the pulse shape, and the packet format. Hence, if this information is not accurate, then the matched filter performs poorly. However, since most wireless network systems have pilot, preambles, synchronization word or spreading codes, these can be used for the coherent detection.
2.2.2 Cyclostationary Feature Detection
An alternative detection method is the cyclostationary feature detection. Modulated signals are in general coupled with sine wave carriers, pulse trains, repeating spreading, hopping sequences, or cyclic prefixes, which result in built- in periodicity. These modulated signals are characterized as cyclostationarity since their mean and autocorrelation exhibit periodicity. These features are detected by analyzing a spectral correlation function. The main advantage of the spectral correlation function is that it differentiates the noise energy from modulated signal energy. Therefore, a cyclostationary feature detector can perform better than the energy detector in discriminating against noise due to its robustness to the uncertainty in noise power. However it is computationally complex and requires significantly long observation time.
2.2.3 Energy detection
Energy detector is simple to design and facilitates the use of low complexity hardware. It is a very generic sensing method as it does not need any knowledge on the primary users’ signal. The signal is detected by comparing the output of the energy detector with a threshold defined by noise energy. The energy detector can detect power levels at certain frequency bands, however it has no means to distinguish between signals from different systems or differentiate interference from a primary user signal and noise. It can only tell whether energy on a signal band exceeds an estimate of noise energy. To achieve good detection performance, the noise power level has to be known accurately a priori which is difficult to achieve in practice. Energy detectors are most suitable for making coarse estimates on channel usage or working side by side with other more advanced detection methods.
Let H0 and H1 represent absence and presence of primary user respectively.
Conventional energy detector uses single threshold (λ) to determine the presence (local decision 1 ) or absence (local decision 0 ) of the primary signal as shown in figure 2.1, where Oi represents energy received by ith secondary user. If energy received (Oi) is greater than threshold (λ), detector will decide 1 and if energy received (Oi) is less than threshold (λ), and detector will decide 0 as shown in equation (2.1):
illustration not visible in this excerpt
Figure 2.1: Conventional Energy Detection with Single Threshold
2.2.4 Autocorrelation-based detection
Autocorrelation based Detector is based on the knowledge of statistical distribution of the autocorrelation coefficient of the received signal. Detection statistic here is time delay value (Td), which is non zero if signal is present and zero if signal is not present. Once the value of the autocorrelation coefficient is computed, the decision can be performed so that a pre-defined false alarm rate specification of detection is fulfilled. Autocorrelation-based detection can effectively detect PU signals under the noise floor. In Autocorrelation based detectors no FFT is done on input signals which makes implementation bit easier but limits the detection to baseband frequencies only.
2.2.5 Cooperative detection
In order to increase detection performance, cognitive radios can also share sensing information with each other. This is called cooperative detection. In Cooperative detection problems due to noise uncertainty, fading and shadowing increases. There are three scenarios for cooperative sensing:
- Centralized sensing: It is like a master slave system in which, a central unit or master collects all the sensing information from the cognitive radios. This Central unit identifies the spectrum occupancy and analyzes the data. Finally, the unit sends the information to the cognitive radios or controls the cognitive traffic directly. The data can be collected as hard or soft decisions depending on the central unit. The hard decision is a binary result that tells whether a PU is present or not. The soft decision is a more accurate test specific data which generally require significant portion of the bandwidth. Therefore when lot of cognitive radios are already present in the spectrum, care has to be taken to use either hard decisions8 or limit the soft decision accuracy accordingly to reduce the bandwidth required to transmit the decisions between cooperating radios.
- Distributed sensing: In this type of scenario, cognitive radios share the sensing information among each other while are independent to take decisions whether to utilize a spectrum band or not. Compared to centralize sensing this method is easier in practice since no central infrastructure is required.
- External sensing: In this an external network performs spectrum sensing and then shares the spectrum occupancy information with the cognitive radios. The external sensor network solves the hidden primary user problem and reduces the performance loss due to fading and shadowing9. External sensor network architectures can be versatile need not to be mobile, reducing problems with power consumption. Also the sensor network, or a single sensor, could sense the spectrum continuously compared to a cognitive radio, which can only sense the spectrum for a short while between the data transmissions.
2.3 Literature Survey
Literature survey is done to find out best method to use for designing the sensing model in this project. In communication environment signal detection and estimation theory is very important to understand to know how to detect the presence of signal. As we have seen that we will model our problem as simple hypothesis testing problem so in this subsection, we first introduce two binary hypothesis testing schemes commonly used in spectrum sensing. Then we discuss composite hypothesis testing methods and sequential testing methods such as the sequential probability ratio test (SPRT) and then we will see how to truncate SPRT with proper bounds is TSPRT. In spectrum sensing, statistical hypothesis testing is typically performed to test the sensing results for the binary decision on the presence of PUs.
In order to take a decision to determine whether a primary user is using the spectrum or not, a statistical model is needed for the PU signals for the detection, and then considers the situation without PU. Detection involves basically a binary (i.e., two-decision) choice between the hypotheses:
H0: signal is absent (only noise is present)
H1: signal as well as noise is present.
The parameter for which the test is carried out will be constrained to the input SNR, which will be constant during the entire test. The particular value of the
input signal-to-noise ratio will be denoted by a. For the case where signal is
present will have some specified value while for the contrary it will be zero, then, has to distinguish between the hypothesis H0 and the alternative hypothesis H1 that a has some specified value a.
Let us assume a simple received signal is modeled as:
illustration not visible in this excerpt
- s(n) is the signal under detection follows the Gaussian distribution with mean a and variance σs, ~Ɲ (μ, σs2)
- w(n) is the additive white Gaussian noise (AWGN) is considered as Gaussian noise with zero mean and variance σn, ~Ɲ (0, σn2)
- n is the sample index
When primary user’s signal is not present then s(n) = 0.
The two hypotheses taken into considerations are such that:
illustration not visible in this excerpt
H0 is a null hypothesis, which states that there is no primary user signal in a certain spectrum band. On the other hand, H1 is an alternative hypothesis, which indicates that there exists some primary user signal.
The detection performance can be expressed with following probabilities:
- Pd: Probability of detection means probability of detecting a signal while signal exits in reality.