CALIFORNIA STATE UNIVERSITY, NORTHRIDGE AUTOMATIC REAL-TIME SPECTRUM SENSING USING ENERGY DETECTION IN SOFTWARE DEFINED RADIO

Transcription

1 CALIFORNIA STATE UNIVERSITY, NORTHRIDGE AUTOMATIC REAL-TIME SPECTRUM SENSING USING ENERGY DETECTION IN SOFTWARE DEFINED RADIO A graduate project submitted in partial fulfillment of the requirements For the degree of Master of Science in Electrical Engineering By Jyh-Chyuan Sun May 2012

2 The graduate project of Jyh-Chyuan Sun is approved: Professor James Flynn Date Dr. Xiyi Hang Date Dr. Sharlene Katz, Chair Date California State University, Northridge ii

3 DEDICATION To my family, for letting go of their dreams so I could pursue a better future. iii

4 ACKNOWLEDGEMENTS I would like to express my deepest gratitude to Dr. Sharlene Katz, Professor James Flynn, and Dr. David Schwartz for their guidance throughout my educational career at California State University, Northridge. The genuine interests they show in engineering are an inspiration to all students at the university, and I know I would not be in the fortunate position that I am in today without them. Their wisdom and guidance made all of this possible, and I am forever grateful for the opportunity to work with them for both my undergraduate group project as well as this graduate project. A special thanks to Dr. Xiyi Hang for serving on my committee and providing insights in MATLAB programming. I sincerely appreciate the candid feedback. Lastly, I would also like to recognize the rest of the faculty and staff in the Electrical and Computer Engineering Department for always being available and ready to assist me in my educational endeavors. iv

5 TABLE OF CONTENTS Signature Page... ii Dedication... iii Acknowledgements... iv List of Tables... vii List of Figures... viii Abstract...x Chapter 1: Introduction History of Software Defined Radio Background: Spectrum Sensing Project Overview...3 Chapter 2: Technical Overview Overview Periodogram Threshold Detection Peak Signal Detection...9 Chapter 3: Hardware Universal Software Radio Peripheral Daughterboard: Wide Bandwidth Transceiver Agilent 33250A...16 Chapter 4: Software MATLAB...17 Chapter 5: Spectrum Sensing...19 v

6 5.1 Overview Graphical User Interface Simulink Data Input Periodogram Threshold Detection Peak Signal Detection Results Display...38 Chapter 6: Discussion of Test Results Results for Amplitude Modulation Results for Frequency-Shift Keying System Limitations...49 Chapter 7: Conclusion Summary Future Work...50 Bibliography...52 Appendix A:...53 Appendix B:...60 Appendix C:...62 vi

7 LIST OF TABLES 3.1 USRP N210 Technical Specifications WBX Daughterboard Technical Specifications Agilent 33250A Technical Specifications Default Test Conditions for Amplitude Modulation Default Test Conditions for Frequency-Shift Keying...46 vii

8 LIST OF FIGURES 2.1 Block Diagram for Main Algorithms Normal FFT (Top) versus Periodogram (Bottom) Band-Pass Filter in Spectrum Sensing Traditional Receiver versus Software Defined Radio Receiver Universal Software Radio Peripheral N Wide Bandwidth Transceiver (WBX) Spectrum Sensing Overall Flow Chart Main GUI Window Program Code for Starting Additional MATLAB Workers Top Level Simulink Model File Inside of the USRP Receiver Block Enabled Subsystem Program Code for Sorting Data into Segments Periodogram Code Task Manager Showing Main MATLAB Application and Four Workers Program Code for Frequency Axis and Power Spectral Density Creation Program Code for Threshold Detection Program Code for Initializing Peak Signal Detection Program Code for Determining Peak Signal Program Code for Recording Possible Peak values Program Code for GUI Graphing and Updating Amplitude Modulation of 10 khz...41 viii

9 6.2 Amplitude Modulation of 10 khz and Carrier of 75 MHz Amplitude Modulation of 10 khz and Display Bandwidth of 20 khz Amplitude Modulation of 10 khz and Resolution Bandwidth of 100 Hz Amplitude Modulation of 10 khz and Resolution Segment of Amplitude Modulation of 10 khz and Pfa of Amplitude Modulation of 10 khz and Baseband of 20 khz FSK Modulation with Resolution Segment of FSK Modulation with Resolution Segment of FSK Modulation with Resolution Bandwidth of ix

10 ABSTRACT AUTOMATIC REAL-TIME SPECTRUM SENSING USING ENERGY DETECTION IN SOFTWARE DEFINED RADIO By Jyh-Chyuan Sun Master of Science in Electrical Engineering The purpose of this graduate project is to design a real-time spectrum sensing system using software defined radio. The uniqueness of software defined radio is the concept of replacing many of the hardware components in a traditional radio communication system with software algorithms and coding. The scope of this project will be on spectrum sensing to be able to automatically detect active signals in the desired frequency spectrum. The hardware components used in this graduate project are the Universal Software Radio Peripheral and the Agilent function generator. The graphical user interface and algorithm programming are performed in MATLAB. The spectrum sensing system will scan a portion of the frequency spectrum, determine the presence of signals, and display the three highest signal peaks in a given band. This paper will elucidate spectrum sensing s strengths and weaknesses as well as possible future work. x

11 CHAPTER 1 Introduction 1.1 History of Software Defined Radio Traditional radio communication systems are typically hardware in nature. The components include an antenna, local oscillator, intermediate frequency (IF) amplifier, demodulator, and an output medium such as a graphic display or audio speakers. There are several disadvantages to having these components in hardware: First, the system takes up space. This is especially important in military applications where the payload is extremely critical. Second, once the system is designed and built, it is essentially permanent. This makes updates and repairs cumbersome and in some cases, impractical. Third, a hardware radio communication system will need to be calibrated and tuned due to drifting issues as well as being susceptible to temperature sensitivity. As a result, this can make the hardware system unreliable at times. The advancement in technology in recent years, however, has allowed the functionalities of these components to be shifted into the personal computer. Collectively, these advancements brought forth software defined radio. A software defined radio communication system places much of the signal manipulation into the digital domain where the computer is then able to process the data. [1] The disadvantages of traditional hardware radios are addressed in software defined radio systems. Space: Software defined radio, by definition, is a reduction in hardware via 1

12 software processing, and a decrease in physical space requirement results. Flexibility: Due to performing the signal processing in software, adding new modes, modulation schemes, or fixes are much simpler than in hardware counterpart. [2] These changes can be done via a software update without the user having to modify the hardware. Reliability: Minimizing the number of hardware components means fewer problems with drifting due to temperature change or the aging of hardware components. 1.2 Background: Spectrum Sensing One application within the world of software defined radio is spectrum sensing. Spectrum sensing is part of a bigger system called cognitive radio, which is a radio that can modify various parameters autonomously for communication purposes without the need for user intervention. Spectrum sensing is the first stage in cognitive radio in which the radio scans a section of the frequency spectrum for any active signal. There are two main applications of spectrum sensing: Scanning for white space, and scanning for signals. These two applications are closely associated with each other. The former is typically for a transmitter design where the system looks at a range of frequency spectrum and uses internal algorithm to decide if there is any white space, and if so, where is it in the spectrum. This allows the system to transmit without interfering another existing signal at the same frequency. The latter is more commonly found in a receiver system where the purpose is to detect any active signal. These two parts form a complete transceiver system. The second application of scanning for active signals is the focus of this graduate project. 2

13 1.3 Project Overview The main goal of this graduate project is to design an automatic software receiver system that will scan a section of the frequency spectrum to look for active signals. First, the user will be able to input specific parameters into a graphical user interface in the computer. The software will adjust the system parameters based on the user inputs. Second, the software will scan the frequency spectrum and analyze for signals. Third, the software will use the graphical user interface to present the results to the user. Chapter 2 will expound the hardware equipment used throughout this graduate project. Chapter 3 will elaborate on the software components and the reasons those components are used. In Chapter 4, in-depth analyses of each portion of the spectrum sensing system will be explained. Discussion on the test results will be in Chapter 5, and Chapter 6 will provide a summary for the work done in this graduate project. 3

14 CHAPTER 2 Technical Overview 2.1 Overview The goal of this spectrum sensing system is to detect and display signals that are present in a specified section of the spectrum by determining if there is energy above the noise in a particular band. From reading the incoming data to displaying the results on the graphical user interface (GUI), there are three main algorithms required: Periodogram, Neyman-Pearson threshold detection, and peak signal detection. A general block diagram of the system is shown in Figure 2.1. The periodogram analyzes the incoming data in order to estimate the power spectrum. The threshold detection algorithm takes the output of the periodogram and calculates a discrete boundary line between active signals and noise based on the variance of the data set from the periodogram. Finally, the peak signal detection algorithm analyzes the data that exceeds the threshold level and returns the three top peaks back to the GUI. The resolution segment, probability of false alarm, and resolution bandwidth are parameters set by the user in the GUI, and these parameters are used by the system for signal processing. In Chapter 5, the code used to implement each of these algorithms is discussed in detail. 4

15 Input Data Periodogram Threshold Detection Peak Signal Detection GUI Resolution Segment Probability of False Alarm Resolution Bandwidth Figure 2.1: Block Diagram for Main Algorithms 2.2 Periodogram The Fast Fourier Transform (FFT) is an efficient method for transforming data from the time domain to the frequency domain. The periodogram is based on the Fourier transform and most often the Fast Fourier Transform (FFT), which is an efficient way of calculating the Discrete Fourier Transform. [3] The difference between the two is that the periodogram takes the FFT of evenly spaced segments of the data rather than the entire data at once. The equation for a periodogram is given as the following: ( ) ( ) (2.1) In the application of spectrum sensing, the periodogram method is superior because it provides a better variance for the set of input data. Variance represents how far apart a particular set of data is spread out in amplitude. [4] The variance of the result of the entire FFT data set does not accurately reflect how far away the individual data points are from 5

16 their mean value. Typically, the variance of the entire FFT data will be larger than the FFT of the data in the segments due to the larger data variations in the entire frame versus the variance of the segments. Because of this, a periodogram will generally produce a smoother graph and enables the system to detect and display signals in the presence of noise. Figure 2.2 shows the difference between a normal FFT and a periodogram. The power spectral density shows the distribution of the total power over a specific range of frequencies. [5] The power spectral density is the Fourier transform of the autocorrelation function, and defined in equation (2.2). [6] However, assuming white noise, the autocorrelation function R xx, reduces to an impulse. [6] ( [ ] [ ] (2.2) [ ] (2.3) A two-sided power spectral density will have a mirror image at the negative frequency. Equation (2.5) is the two-sided power spectral density, with equation (2.4) as angular frequency ω measured in radians per second. (2.4) ( ) [ ] (2.5) 6

17 Figure 2.2: Normal FFT (Top) Versus Periodogram (Bottom) 7

18 The graduate project, however, will only utilize a one-sided power spectral density. The reason is to simplify the peak signal detection algorithm and prevent the possibility of counting the same peak twice. 2.3 Threshold Detection The type of hypothesis testing used is the Neyman-Pearson threshold detection. [7] This threshold is defined in equation (2.6). ( ) ( ( ( )) ) (2.6) The variable λ is the threshold and σ 2 is the variance of the amplitude data. The function erfc -1 is the inverse of the complementary error function. The complementary error function is one minus the error function, which is related to the normalized Gaussian function. [8] The Gaussian function is a probability distribution, and has the classic bellshaped curve. The variable P FA is the probability of false alarm, which is a parameter that is set by the user in the GUI. The variance is calculated internally by the system with a given set of data. Equations (2.7) and (2.8) define the complementary error function and the normalized Gaussian function, respectively. [8] ( ) ( ) (2.7) ( ) (2.8) 8

19 By setting a desired probability of false alarm and calculating the variance of a data set, the system sets a threshold to indicate signals above the noise level. 2.4 Peak Signal Detection The peak signal detection algorithm determines possible peak signals in the scanned spectrum. The algorithm uses the concept of band-pass filtering. In a simple RC case, a low-pass filter cascaded with a high-pass filter will result in a band-pass filter. Equation (2.9) is the general cutoff frequency used for both low-pass and high-pass filters. [9] (2.9) In equation (2.9), the variable f c is the cutoff frequency, which is 3 db down from the filter s center frequency. The center frequency, or resonant frequency, can be represented in equation (2.10). (2.10) Even though a software solution has no resistor or capacitor, the same concept can be applied. By storing the amplitude values in an array, the peak value can be easily discerned. In this software solution, the difficulty lies not in determining peaks, but to be able to distinguish one signal s peak from noise as well as other signals. The assumption here is that the both the lower cutoff frequency and the higher cutoff frequency are equidistant from the resonant frequency. The bandwidth is the range between the lower 9

20 cutoff frequency and the higher cutoff frequency. This bandwidth is a known value because it is the resolution bandwidth parameter set by the user in the GUI. The equation for the bandwidth is shown in equation (2.11). [9] (2.11) The algorithm will place the midpoint of the bandwidth over the peak value. Data points within the bandwidth will be considered as the same signal and passed through. Figure 2.3 shows an example of band-pass filtering applied to the spectrum sensing application. Figure 2.3: Band-pass Filter in Spectrum Sensing 10

21 Using Figure 2.3 above, assume an f r of 20 khz and the user sets a bandwidth of 10 khz. The middle of the bandwidth will be placed at 20 khz, and the bandwidth will span 5 khz to the right and to the left of 20 khz. Data points below the threshold level are treated as noise and not included in the peak signal detection algorithm. The algorithm searches through the data points to find the first peak value. It considers all data points inside that band and above the threshold level (the red line in Figure 2.3) as from the same 20 khz peak signal. The algorithm will remove the peak value at 20 khz and all data points inside the bandwidth, and restart the process to see if there is another peak signal above the threshold. Finally, the algorithm will return a maximum of three peak values back to the GUI. In this application, it is assumed that the peak value is at or very close to the center of the signal or the bandwidth placement might be skewed to the left or the right potentially rejecting part of its own signal or passing data points from a neighboring signal. The filtering enables the algorithm to disregard peak values outside of the bandwidth and only analyze the values within the bandwidth. 11

22 CHAPTER 3 Hardware 3.1 Universal Software Radio Peripheral While many of the components in a software defined radio are shifted to the software side, some hardware components are still needed. Figure 3.1 below shows the difference between a traditional radio and a software defined radio. Traditional / Hardware Receiver RF Amplifier x IF Amplifier Demodulator Local Oscillator Software Defined Radio Receiver Receiver Front End ADC Software Figure 3.1: Traditional Receiver vs. Software Defined Radio Receiver The hardware side has more than one option: In the test instrument industry, there is an active focus to move towards PXI, or PCI extensions for Instrumentation. Originally introduced by National Instruments in 1997, PXI modules are slot-based card solutions for traditional hardware instruments like spectrum analyzers and oscilloscopes. [10] Another system, and the one used in this graduate project, is Universal Software Radio Peripheral, or USRP for short. USRP is a low-cost solution for a software defined radio 12

23 system. There is a set of generic hardware on the USRP, such as analog-to-digital converters (ADC) and digital-to-analog converters (DAC). The rest of the work is done on the computer. Figure 3.1: Universal Software Radio Peripheral (USRP) N210 [11] The USRP has a set of basic hardware components. Table 3.1 below shows the specifications for the N210 USRP: USRP N210 Specifications FPGA: Xilinx Spartan 3A-DSP 3400 Dual ADC: 100 MSamples/s Dual DAC: 400 MSamples/s Connectivity to PC: Gigabit Ethernet Table 3.1: USRP N210 Technical Specifications [12] The radio frequency (RF) range of the USRP varies depending on the daughterboard used. This is a modular approach and mimics the slot-based card solutions aforementioned in the PXI system. The onboard ADC converts analog signals coming 13

24 over the air into digital representations for the computer to process. The DAC reverses this process and converts the digital data from the computer into analog signals to transmit over the air. Due to the massive amount of data streamed to the computer, the USRP N210 requires gigabit Ethernet or the computer will not be able to communicate with the USRP. Through experimentation, it is believed this limitation is hardcoded into the USRP firmware to ensure the system will not slow down due to 10/100 Ethernet speed. However, one workaround for computers that do not have gigabit local area network (LAN) is to use a gigabit switch. Because the USRP is connected to the gigabit switch, the former believes it has connected to a computer with gigabit LAN. The gigabit switch then connects to the computer without gigabit LAN. Although this will allow the computer with non-gigabit LAN to communicate to the USRP, it is not recommended due to potential overrun and underrun problems. Overrun occurs when the computer is not keeping up with the incoming data from the USRP. [13] Conversely, underrun is when the computer is not transmitting the data to the USRP fast enough.[13] This is the reason for the USRP requiring gigabit LAN speed to minimize overruns and underruns. 3.2 Daughterboard: Wide Bandwidth Transceiver There are many modular daughterboards that can be used on the USRP depending on the RF ranges desired. This graduate project uses the wide bandwidth transceiver, or WBX daughterboard. 14

25 Figure 3.2: Wide Bandwidth Transceiver (WBX) [14] The WBX daughterboard covers a range of frequencies from 50 MHz to 2.2 GHz, but the frequency that will be mainly used in this graduate project is 70 MHz to 75 MHz. Although the minimum frequency of the WBX daughterboard is 50 MHz, experiments showed that there was distortion and nonlinearity near the WBX s minimum frequency. For this reason, the targeted frequency range for testing purposes is 70 MHz to 75 MHz. The technical specifications for the USRP WBX daughterboard are provided in Table 3.2 below. WBX Daughterboard Minimum Frequency: 50 MHz Maximum Frequency: 2.2 GHz Maximum Output Power: 100 mw Noise Figure: 5 db Bandwidth: 40 MHz Table 3.2: WBX Daughterboard Technical Specifications [15] 15

26 3.3 Agilent 33250A The input test signal used in this graduate project is from the Agilent 33250A Function/Arbitrary Waveform Generator. The generator is readily available in the laboratory and Table 3.3 provides the specifications for the instrument A Function/Arbitrary Waveform Generator Modulation: AM, FM, FSK AM Carrier: Sine, square, ramp, and arb AM Modulated Frequency: 2 mhz to 20 khz Sine: 1 uhz to 80 MHz Square: 1 uhz to 80 MHz Pulse: 500 uhz to 50 MHz Filter Bandwidth: 50 MHz Amplitude Resolution: 12 bits Sample Rate: 200 MSamples/s Table 3.3: Agilent 33250A Technical Specifications [16] Because the Agilent 33250A has a maximum frequency of 80 MHz for sine wave generation, tests were performed in the 70 MHz to 75 MHz range to prevent distortion and nonlinearity from the generation side. To minimize distortion caused by the 33250A, a decision was made to lower the maximum test frequency to 75 MHz. 16

27 CHAPTER 4 Software 4.1 MATLAB The software used in this graduate project is called MATLAB. Produced by MathWorks, MATLAB offers a programming environment for developing algorithms, data processing, and graphical user interface creation. [17] MATLAB, in the case of this graduate project, offers the complete package on the software side. Furthermore, MATLAB is capable of communication with the USRP, eliminating the need to manually implement a protocol pipeline to communicate with the USRP. MATLAB is a compiler, meaning that it can compile written code immediately in a similar fashion to Microsoft Visual Studio. This advantage means prototyping can quickly be built, tested, and implemented. Alternatively, another option to interface with the USRP hardware is GNU Radio and GNU Radio Companion (GRC). GNU Radio is a free and open-source software package capable of signal processing algorithms and graphical visualizations similar functionalities as provided by MATLAB. [18] GNU Radio mainly operates in the Linux operating system environment. The source code for GNU Radio is written in the programming language Python. GRC is a software toolkit that is bundled with GNU Radio; it is also free and open-source. GRC is a graphical tool, utilizing blocks for implementation at a systems level. [19] GRC allows the user to create a general system template visually and contains a feature to convert the blocks into equivalent Python 17

28 code. Therefore, the main purpose of GRC is to reduce the amount of programming needed via GNU Radio alone. GNU Radio and GRC are not used in this graduate project. MATLAB has very good built-in documentation and the maturity of Simulink a toolbox within MATLAB and acts as the input and output points to communicate with the USRP is sufficient in this project. In addition, Simulink is comparable to GRC; in other words, a graphical block layout tool that allows the user to implement a system without the need to program manually. This is not to say MATLAB has no disadvantage: MATLAB is a large software containing various toolboxes and functionalities. Occurrences of overruns and underruns are more prevalent in MATLAB, thus the reason for gigabit LAN is so crucial. 18

29 CHAPTER 5 Spectrum Sensing 5.1 Overview At the system level, the GUI is the central point for all stages in spectrum sensing. The GUI has a set of basic parameters for the user to change; it acts as both the input and the output interface for this system. The input refers to the user input and the output is where the data results are displayed. After the GUI reads the user inputs, it starts the Simulink toolbox within MATLAB. Simulink is used to allow the program to communicate with the USRP hardware. Simulink takes the user inputs and adjusts the USRP accordingly. The USRP will tune to the desired frequency and stream the data back to the main MATLAB interface. It is worth noting that the USRP is not a smart radio by itself: The intelligence in a SDR is due to the program code. After receiving and storing the data in Simulink, the program code analyzes the data in order to determine the presence of signals in the scanned frequency spectrum. The top three peak signals report back GUI and the GUI displays the result. The process repeats indefinitely or until the user manually stops the system. The implementation of spectrum sensing in MATLAB using USRP has a general flow chart as Figure 5.1 illustrates. The algorithm processing block is a multistep process performed in MATLAB, and the main topic of this graduate project. 19

30 Graphical User Interface Read User Inputs Simulink Read USRP Inputs Algorithm Processing Display Results Figure 5.1: Spectrum Sensing Overall Flow Chart 20

31 5.2 Graphical User Interface The GUI in MATLAB is designed using Guide, which stands for GUI Development. The GUI has a simple and easy-to-use interface available to the user. When the program first starts, the GUI appears as shown in Figure 5.2. On the left-hand side, there is a visual frequency spectrum display. The vertical axis is the signal amplitude in decibels (db) and the horizontal axis is the frequency in hertz (Hz). The screen starts out with a blank graph as shown in Figure 5.2. Below the graph, there are two buttons: The Start button initiates the spectrum sensing and analysis, and the stop button manually halts the scanning process. Because this is a continuous system that runs indefinitely, as long as the user does not explicitly press stop, the system will continue running. Figure 5.2: Main GUI Window 21

32 The right-hand side of the GUI is split into two panels. The top panel is the control panel, and in this panel all the parameters are changeable by the user. The center frequency is a parameter required in the Simulink USRP block. This parameter sets the center frequency of the frequency scan. The display bandwidth adjusts the horizontal axis in the frequency spectrum display. This flexibility allows the user to look at a wide frequency range as well as being able to focus and zoom in on a narrow frequency region. Resolution bandwidth determines whether data points are part of the same signal. Higher resolution bandwidth results in faster processing, but the output result is not as clear. [20] As it pertains to this graduate project, resolution bandwidth is used for the peak signal detection algorithm. Resolution bandwidth is needed because the peak signal detection algorithm cannot visually discern if data points are from the same signal, multiple signals nearby, or noise. For example, suppose there is a signal with a peak of 70 MHz. If the resolution bandwidth parameter is set at 2 MHz, then the algorithm will come to the conclusion that all data points within 69 MHz to 71 MHz are the same signal. The user would either need to know the desired signal type or adjust this parameter based on the spectrum graph in the GUI. The resolution segment parameter is used at various points throughout the system. Resolution segment helps determine the fast Fourier transform (FFT) length. Smaller resolution segment results in more samples when performing FFT. While this means better clarity in the frequency display, it does come with more acquisition and 22

33 computation time. Because the resolution segment is used for determining FFT sample points, powers of two are strongly recommended such as 256, 512, or The equation for calculating the FFT length is shown in equation (5.1). (5.1) The system designed in this project uses a frame length of Therefore, using equation (5.1) with a resolution segment of 64: The Pfa parameter stands for the probability of false alarm, and used in the Neyman- Pearson threshold detection. It is a statistical metric used for threshold detection. Setting the probability of false too small might result in false negatives and setting the value too high may return false positives. When the probability of false alarm is set too high, the threshold level for determining whether a signal is active or not is lowered in a reciprocal fashion. If the threshold level is set very low, for example, at the noise floor, essentially all data samples are considered active signals creating many false negative results and rendering the spectrum sensing system useless. 23

34 Below the control panel is the display panel. The display panel will show the peak amplitude signal with its corresponding frequency value to the user for each detected signal. If no active signal is detected, both text boxes will display Lastly, there are two notes at the bottom of the display panel to remind the user to set the resolution segment in powers of two, and -999 in the boxes means no signal was found. When the GUI is first started, the input boxes are populated with specific default values. All of these values may be adjusted, but nothing in the display panel can be changed since they are meant for display purposes only. When the user clicks on the start button, the static text to the right of the Start and Stop buttons will be changed from STOPPED to Processing Data. This informs the user that the system is readying itself to scan the spectrum. The variables instantiated in MATLAB cannot be accessed or are they even visible in the variable workspace in Simulink. This resulted in difficulty when trying to pass variables between MATLAB and Simulink, such as the center frequency. The solution is to use the command assignin, which stands for assign in. As the name suggests, the command assigns variables into the workspace. However, this by itself does not create a common workspace between MATLAB and Simulink. The workspace parameter in the command must be explicitly set to base, as that is the common workspace shared by MATLAB and all derivatives of MATLAB including Simulink. 24

35 Although the input boxes in the control panel look like numbers, they are actually all strings. Therefore, the code must first convert all the string values into numerical numbers (in the double format) so MATLAB can take these values and do further processing. The MATLAB GUI in general communicates in handles. For instance, the spectrum display is named handles.main_graph, and the center frequency string is handles.edit_cf. In in the initialization, the program code will check to see if multiple workers are started. Workers act as additional processes in the Windows operating system, and MATLAB is able to take advantage of multiprocessing if paired with the parfor-loop, which will be discussed in section 5.4. The code in Figure 5.3 will query the how many additional workers are started. If the result is zero, the code will generate four additional workers since the computer used has a quad-core central processing unit (CPU). The full MATLAB program code m-files are located in Appendices A, B, and C. Figure 5.3: Program Code for Starting Additional MATLAB Workers 25

36 After the conversion is complete, the program will run Periodogram.m, a MATLAB m- file. At the top of Periodogram.m, the rest of the input string parameters are converted into double precision values. After this initialization process, Periodogram.m calls the Simulink model file. 5.3 Simulink Data Input At the top level of this Simulink model file, there are only three blocks: a constant source block, the USRP receiver block, and an enabled subsystem block. The USRP block requires an input value for the center frequency. This is the input value that the user sets in the GUI as described in section 5.2. The constant source block simin_input_signal takes the user carrier frequency input value from the GUI and sends it to the USRP block. The command assignin allows Simulink to access and read the center frequency value through a common workspace. The USRP receiver block is denoted as SDRu Receiver in Figure 5.4 below. The USRP communicates via the Ethernet LAN port. The static internet protocol (IP) address set by the USRP firmware is

37 Figure 5.4: Top Level Simulink Model File The internal parameters for the USRP receiver block are shown in Figure 5.5. The center frequency parameter does not contain an input box. This is because the center frequency value is controlled by the constant source block. The rest of the values are default settings that should be left alone. From experimentation, these values are set at a gain of 10 db, decimation of 500, sampling time of 5 x 10-6 seconds, and frame length of The onboard digital downconverter (DDC) decimates the input signal from a higher sampling rate of 100 MSamples/s in the field programmable gate array (FPGA). Using equation (5.2) below, signal processing in the computer occurs at a lower sampling rate of 200 khz. (5.2) 27

38 The USRP receiver block in Figure 5.5, however, expects sampling time, not sampling frequency. Sampling time has an inverse relationship to sampling frequency, and is shown in equation (5.3). (5.3) The data coming out of the USRP receiver block is fed into the input of the enabled subsystem, and the data length is connected to the enable block inside the enabled subsystem. This tells the subsystem to run only when a frame is received. Inside of the enabled subsystem is where the actual signal processing takes place. This is shown in Figure 5.6. This subsystem sends the incoming signal to the workspace variable called simout_input_signal. Because there is an integer delay, the variable simout_input_signal in the Simulink workspace will have multiple rows of length

39 Figure 5.5: Inside of the USRP Receiver Block 29

40 Figure 5.6: Enabled Subsystem Through experimentation, it was discovered that the USRP initially buffers erroneous data. The exact cause for this is unknown. Setting the delay to three will actually generate four rows of data because the matrix contains one row of data without delay, and three rows of data with delays. Therefore, only the last frame is used while the previous ones are discarded. The integer delay could be set longer, and in fact, has been tested with 10 and 20. However, not only did this increase the processing time, it did not increase the accuracy or fidelity of the signal. Through testing, a delay of three was found to yield good data while minimizing the chance of buffering erroneous data. The data type conversion block is used after the delay block to convert its output to the same signal type as the stop simulation block. The stop simulation block informs the subsystem to stop recording data coming from the USRP. 30

41 Once Simulink has stopped recording data points, it returns control back to Periodogram.m for the next step in the system process. It should be noted that MATLAB has access to simout_input_signal, or the Simulink output of the input signal from the USRP. 5.4 Periodogram When Periodogram.m has control of the program once again (it pauses while the Simulink model file is still active), it processes the last frame of the data that was received. The next step in the process is shown in Figure 5.7. The code takes the data points into multiple rows based on the resolution segment set by the user in the GUI. The purpose for doing this is to improve processing time. Figure 5.7: Program Code for Sorting Data into Segments After organizing the data into the desired segments, the next process is to compute the periodogram. In the program code, the periodogram is found by taking the FFT of a segment. The algorithm can be represented in Figure 5.8. The resolution segment parameter set by the user in the GUI helps determine the number of FFT points. Recall the discussion in section 5.2 that smaller resolution segments will result in more data 31

42 points and thus finer resolution. In Figure 5.8, the resolution segment, rseg in the program code, is on the equation s denominator. Therefore, a smaller resolution segment will result in a higher number of points for the FFT. The periodogram uses the special for-loop called parfor. The parfor-loop is a parallel for-loop meant to take advantage of a multicore computer, and increases speed versus traditional for-loops. Parfor-loops exist in MATLAB s Parallel Computing Toolbox. Figure 5.8: Periodogram Code It should be noted, however, that the parfor-loop cannot be used in all circumstances. This is due to MATLAB s restriction on executing parfor-loops when there are additional nested loops within. MATLAB internally sends portions of the data to be computed in the parfor-loop to multiple workers (separate processes at the operating system level), but with nested loops the workers may work asynchronously due to the nature of nested loops. There must be only one for-loop so MATLAB implicitly knows how many computations there are, and splits the workload among its workers evenly. The initialization of workers is done before the GUI is visible, and was discussed in section 32

43 5.2. Figure 5.9 shows MATLAB with four additional workers to offload the data from the main MATLAB process. Figure 5.9: Task Manager Showing Main MATLAB Application and Four Workers After the periodogram is computed, the following process creates the frequency axis for plotting purposes and finds the one-sided power spectral density. The program code in Figure 5.10 shows this process. Figure 5.10: Program Code for Frequency Axis and Power Spectral Density Creation The command floor in MATLAB rounds to the nearest lower integer and causes the horizontal axis to have integer values. 33

44 5.5 Threshold Detection Data points above, or passing, the threshold count as active signals in the peak signal detection algorithm. Conversely, data points below the threshold will be viewed as noise. The threshold is really nothing more than a numeric limit separating the active signals from the noise. The MATLAB program code for threshold detection is shown in Figure Figure 5.11: Program Code for Threshold Detection The MATLAB command npwgnthresh is a Neyman-Pearson threshold used in hypothesis testing for threshold detection. [7] The npwgnthresh command uses the probability of false alarm set by the user in the main GUI window. The result is the signal-to-noise (SNR) threshold in decibels. 5.6 Peak Signal Detection The final stage in the data processing is the peak signal detection algorithm. This algorithm is the decision engine that takes the data from periodogram along with the threshold level and puts them together to determine the presence of signals. 34

45 The program code shown in Figure 5.12 determines if there is any data point above the threshold level. If no data point is found to be above the threshold level, there is no need to even enter the peak signal detection algorithm. This is more efficient, especially if it turns out that there is no signal in the spectrum to begin with. By default, Periodogram.m will return -999 for both the amplitude and its respective frequency back to the main GUI window. This information is used in the last stage of the signal processing, and will be discussed in section 5.7. Figure 5.12: Program Code for Initializing Peak Signal Detection There is an initial check to see if the threshold level calculated is too low. If the threshold level is below -70 db, the system assumes an error must have occurred somewhere (user or otherwise) and returns -999 back to the main GUI window. 35

46 The main program code for the peak signal detection algorithm is shown in Figure The purpose of the program code in Figure 5.13 is to compare and hold the peak amplitude values, if they exist. If the size within lock_f is equal to one exactly, there is only one peak value above the threshold. Therefore, there is no need to do any further data manipulating and simply hold that value. If the size is less than one which means an error has occurred because the spectrum only contains zero and positive values the program code will exit the entire if-loop. Figure 5.13: Program Code for Determining Peak Signal 36

47 If the size is more than one, the program code will remove non-peak values. This is determined via the resolution bandwidth, which acts as a filter based on the peak amplitude value. Data points within the resolution bandwidth of the peak are regarded as the same signal, and thus ignored by the algorithm. The last portion of the code in the peak signal detection algorithm is to populate arrays peak_freq and peak_amp (both contains three elements) with the three top peaks returning to the main GUI window. However, if the number of peak values is less than three, the algorithm will fill the empty array elements with The program code for returning possible peak values is shown in Figure Figure 5.14: Program Code for Recording Possible Peak Values 37

48 5.7 Results Display After the execution of Periodogram.m, the main GUI assumes control once again. The program code shown in Figure 5.15 displays the graph result, draws the threshold level if it is greater than -70 db, and displays the three highest signal peaks in the main GUI window. The code will also readjust the horizontal axis based on the display bandwidth parameter set by the user. Figure 5.15: Program Code for GUI Graphing and Updating 38

49 This reaches the end of the full cycle for the spectrum sensing program. After the graph has been adjusted and plotted, the GUI will continue indefinitely until the user clicks on the stop button. The program will not stop immediately because the check to see whether the stop button has been clicked is located at the beginning of the program run. Alternatively, the program can contain multiple checks at various key points in the program run to see whether the stop button has been clicked or not; however, this will burden the system with even more execution code. In the end, the decision is to avoid unnecessary conditional checks throughout the code and prevent further speed degradation. 39

50 CHAPTER 6 Discussion of Test Results 6.1 Results for Amplitude Modulation As discussed earlier, a realistic target frequency range due to limitations on both the USRP WBX daughterboard and the Agilent 33250A generator is 70 MHz to 75 MHz. In the first test, the generator is set to a carrier frequency of 70 MHz with a baseband amplitude modulation (AM) of 10 khz. The default test conditions are shown in table 6.1, and all other tests in section 6.1 will be compared to this default case. Default Test Conditions Agilent 33250A GUI Carrier Frequency: 70 MHz Center Frequency: MHz Baseband Frequency: 10 khz Display Bandwidth: 50 khz Modulation: AM Resolution Bandwidth: 1 khz Resolution Segment: 256 Pfa: 1e-4 Table 6.1: Default Test Conditions for Amplitude Modulation The signal in Figure 6.1 is clearly visible in the main graph. According to the returned results, the spectrum sensing program found the baseband peak at 11.0 khz with an amplitude level of db. The threshold level is set to approximately -32 db. Because the peak signal is above the threshold, the spectrum sensing program regarded the signal as an active signal. The shifted frequency axis is due to the USRP, so the peak signal centered on zero is actually the 70 MHz carrier. The baseband is at 11 khz because the center frequency parameter is shifted by 1 khz on the control. The display bandwidth is 40

51 set to 50 khz, so the horizontal frequency axis is set to a maximum of 5x10 4 Hz, or 50 khz. Figure 6.1: Amplitude Modulation of 10 khz Figure 6.1 above is regarded as the default condition, and the following tests in section 6.1 will be to see the result based on parameter changes. Figure 6.2 shows the result with the generator changing its carrier frequency to 75 MHz. The center frequency set by the user has an offset of 5 khz, which is reason the baseband frequency shown in the GUI is 15 khz instead of 10 khz. 41

52 Figure 6.2: Amplitude Modulation of 10 khz and Carrier of 75 MHz The next parameter change is the display bandwidth to 20 khz. Referring to Figure 6.3, the baseband signal is visually shifted to the middle of the graph because the maximum horizontal axis is now only 20 khz. Shown in Figure 6.4, raising the resolution bandwidth to 5000 Hz resulted in the removal of the erroneous peak at 2.36 khz that was present in Figure 6.1. Due to the higher resolution bandwidth, the system interprets all data points ± 5000 Hz of the peak at Hz as the same signal and thus will not report the Hz peak back to the GUI. 42

53 Figure 6.3: Amplitude Modulation of 10 khz and Display Bandwidth of 20 khz Figure 6.4: Amplitude Modulation of 10 khz and Resolution Bandwidth of 100 Hz 43

54 Changing the resolution segment from 256 to 1024 will theoretically have the peaks be less distinguishable. As shown in Figure 6.5, this is exactly what happens. The display panel no longer reports back the correct peak of 11 khz. By lowering the resolution segment and thus the FFT length fewer points will be sampled. The result is the displayed peak value is incorrect. Figure 6.5: Amplitude Modulation of 10 khz and Resolution Segment of 1024 The probability of false alarm affects the threshold level. Setting the probability of false alarm to 0.1 lowered the threshold level and introduced false peak values. 44

55 Figure 6.6: Amplitude Modulation of 10 khz and Pfa of 0.1 Another change is to set the baseband to 20 khz, and the result is shown in Figure 6.7. Figure 6.7: Amplitude Modulation of 10 khz and Baseband of 20 khz 45

56 6.2 Results for Frequency-Shift Keying The generator is also able to generate frequency-shift keying (FSK) signals, and the tests in section 6.2 are performed with FSK modulation. The default test conditions are shown in table 6.2. Default Test Conditions Agilent 33250A GUI Carrier Frequency: MHz Center Frequency: 70 MHz Hop Frequency: MHz Display Bandwidth: 50 khz FSK Rate: 3 khz Resolution Bandwidth: 1 khz Modulation: FSK Resolution Segment: 256 Pfa: 1e-4 Table 6.2: Default Test Conditions for Frequency-Shift Keying This time, suppose the only parameter known is the input signal s carrier frequency is at 70 MHz. If this is the case, the default test conditions in the GUI can be the same as section 6.1. Figure 6.8 shows only a peak at 20 khz, but nothing else is very indistinguishable. Therefore, the next step is to lower the resolution segment to 64. Recall from section 4.2 that lowering the resolution segment will increase the FFT length in the periodogram, resulting in more data points and a clearer graph display. Figure 6.9 is the graph display with the resolution segment at

57 Figure 6.8: FSK Modulation with Resolution Segment of 256 Figure 6.9: FSK Modulation with Resolution Segment of 64 47

58 The result is clearer and the sidebands are close to the peak. However, the threshold level changed. This is due to the adjustment in resolution segment, which changes the data set that the system uses to calculate the variance. Another parameter that can be modified is to set the resolution bandwidth to a lower value. The reason the two lower peaks reported back incorrectly is because they are less than 1000 Hz from the top peak. To the system, it is simply concluding that those two values are from the same signal as the one at 20 khz. Decreasing the resolution bandwidth will reintroduce the two peaks ± 1000 Hz of the top peak back into consideration. This is shown in Figure Figure 6.10: FSK Modulation with Resolution Bandwidth of 100 The other two peaks are now showing correctly in the display panel. From the information in Figure 6.10, it is possible to conclude that the hop frequency is 20 khz 48

59 higher than the carrier frequency of 70 MHz. In addition, the FSK rate is set to approximately khz, which is very close to the actual FSK rate of 1 khz. 6.3 System Limitations Although the spectrum sensing system passed several test cases in section 6.1 and 6.2, it is not a perfect system. For instance, the user still needs to read the spectrum display and adjust several parameters in order to see the correct peak amplitudes and frequencies in the GUI. In addition, the system introduced erroneous peak values in some instances, which is due to the resolution bandwidth setting being too high or too low. Currently, the system is unable to automatically adjust the GUI parameters and rescan to see if a better result can be gathered. This leads to the user having to adjust the GUI parameters in order to eliminate erroneous peak values or refine the display. Lastly, the system has not been tested with multi-tone signals or other modulated signals. The behavior of the system for these test cases is unknown. 49

60 CHAPTER 7 Conclusion 7.1 Summary The main purpose of this project is to design a spectrum sensing receiver program using software defined radio. The USRP hardware system defined the scope of the project, and the software program illustrated the potential of what a software defined radio system is capable of doing. The spectrum sensing program contains a main GUI window for the user. There are many parameters that the user can change and set. Once started, the program opens a Simulink model and interfaces with the USRP hardware. The program tells the USRP the center frequency to tune to. The USRP streams the data and saves it in a matrix. The program then initiates the periodogram analysis to not only transform the data to the frequency domain, but determine whether a signal is viewed as an active signal or not. Once processed, the program returns the results back to the main GUI window to the user display. The user is capable of manually adjusting the parameters to fine-tune the details or scan another portion of the spectrum. This system has been shown to perform as expected in a number of test cases. 7.2 Future Work Due to the modular nature of software defined radio, there is potential for future work in this area. The spectrum sensing system is constructed in such a way that additional 50