RESEARCH ON METHODS FOR ANALYZING AND PROCESSING SIGNALS USED BY INTERCEPTION SYSTEMS WITH SPECIAL APPLICATIONS

Transcription

1 Abstract of Doctorate Thesis RESEARCH ON METHODS FOR ANALYZING AND PROCESSING SIGNALS USED BY INTERCEPTION SYSTEMS WITH SPECIAL APPLICATIONS PhD Coordinator: Prof. Dr. Eng. Radu MUNTEANU Author: Radu MITRAN

2 1. INTRODUCTION The objective of this paper is to investigate algorithms used for detecting radio signals used in VHF and UHF terrestrial communications. The VHF and UHF frequency ranges are used mainly for commercial mobile communications and special applications. During research performed, detection methods based on measuring radio signal energy and detection algorithms exploiting signal characteristics were analyzed. Evaluating the performances of the studied algorithms, radio signals used for special applications and for mobile communications were used. Three different test platforms for analyzing signals of interest and practically testing detection algorithms were employed. The usefulness of these algorithms is reflected both in specific applications (signal intelligence, electronic warfare) and in commercial applications. In order to protect access to the electromagnetic spectrum and to selectively block access for unauthorized users, a new specialty called electronic warfare was constituted. At the same time, the need to obtain information by intercepting communications and technically analyzing radio signals led to designing special interception systems. These systems are useful because they perform the following functions: early warning for avoiding surprise; establishing the enemy s positions on the battlefield; target acquisition; selecting and applying countermeasures; extracting information by listening on radio traffic. Special applications use electronic systems for communications, target detection, weapon guidance systems, etc. (Adamy, 2004). Modulations used include analogical modulations (AM, FM) and digital modulations (CPFSK, PSK). Tactical communications high complexity makes it necessary to use robust algorithms for detection, tracking and classification of signals. Due to the high complexity of tactical communications, processing algorithms used a large variety of radio signals. The emergence of digitally modulated signals, spread spectrum techniques (frequency hopping, direct sequence), and burst signals imposed new requirements for interception systems. The complexity of these signals has allowed increased protection against interception. From considerations mentioned above it is necessary to investigate new algorithms used in special interception systems to meet growing complexity and resistance to interception of radio signals used in tactical communications. Research conducted in this paper investigates the ability of algorithms to detect digitally modulated signals (CPFSK) and frequency hopping signals (e.g. TETRA downlink and GSM downlink). In the field of commercial communications, applications monitoring the electromagnetic spectrum appeared only recently. The first applications involving spectrum observation were included in the cellular mobile communication standards for finding the control channel (BCCH). Modern algorithms for studying the electromagnetic spectrum were intensely studied and practically implemented starting with the introduction of the cognitive radio station concept in the early 2000 (Mitola, 1999). The spectrum sensing capacity plays an important role in the research activities performed for developing adaptive radio stations. Spectral sensing algorithms were tested and implemented in a series of radio station types with cognitive radio stations specific capabilities. There are three communication standards including spectrum sensing capabilities and technical specifications. These standards are: k, , and Bluetooth. 2

3 802.11k is an upgrade of and includes measurements used for monitoring free channels and link quality. These measurements generate statistics on the traffic sent on a channel, noise histograms, BER level, interference temperature, received signal level on available channels, etc. Bluetooth introduced a new capability called adaptive frequency hopping (AFH). This algorithm involves sensing the spectrum and avoiding occupied channels in order to reduce the level of interference with other communications systems operating at 2.4 GHz. The detection algorithm is based on statistic information collected about occupied channels in this frequency band uses advanced spectrum sensing capabilities, being the first standard with cognitive radio station specific capabilities. This standard is currently being developed and regulates the operation of wireless rural area networks (WRAN). Primary users of the allocated frequency band are terrestrial TV channels and wireless microphone channels. Theoretical principles of signal detection were displayed in various papers (Trees 2001, 2001, 2002). Signal detection can be modeled in most cases as being a matter testing two hypotheses: the presence or absence of signal. Poison (2004) conducted a survey of the main algorithms used for testing the hypotheses. Performance of these algorithms depends on signal characteristics and radio channel conditions. Since the information signal is send randomly, radio signals used for communications can be modeled as a sine wave with random parameters. When all signal parameters are known (e.g. synchronization sequences, training sequences), the adaptive filter is the optimum test for signal detection. When one or more parameters of the modulated signal are unknown, the adapted filter performances degrade rapidly. In this situation there are algorithms ensuring superior performances for the adapted filter. Traditional detection methods perform spectrum sensing in two dimensions: frequency and time. The first definitions of the electromagnetic spectrum defined it as being tridimensional, with time, frequency and space components. From this perspective, detection algorithms are designed to determine the frequency bands (channels) occupied in bounded geographical areas and defined moments of time. Development of digital communications and new transmission techniques brought new definitions of the electromagnetic spectrum. Matheson (2003) considers the electromagnetic spectrum to be a hyperspace with given frequency, time, space, code, and angle of arrival dimensions. This definition allows precise characterization of spread spectrum signals (hopping codes, spreading codes) and MIMO signals. Changing the definition of the electromagnetic spectrum brought about the occurrence of detection algorithms exploiting all of its five dimensions. 2. THESIS STRUCTURE Chapter 2 contains a technical analysis of the main types of signals used in special applications and of TETRA and GSM downlink signals. Characteristics of radio signals of interest were based on standards and were confirmed by measurements. Measurements were performed in the time domain, frequency domain, and time-frequency domain. Three distinct measurement and analysis systems were used: R&S FSH3 digital spectral analyzer; Narrowband signal acquisition system; Wideband signal acquisition system. The narrowband signal acquisition system uses a 2 MHz 3 GHz radio receiver for translating the RF signal intercepted on an intermediate frequency. This signal is conversed to a digital format by an acquisition board installed on a PCI slot of a PC. MATLAB functions were used for the spectral analysis of the signal. The wideband signal acquisition system is composed of the following elements: 3

4 log periodic antenna for the 10 MHz 3GHz frequency range; Keithley 2820 signal analyzer; PC with MATLAB functions for signal processing. The route of the intercepted signal, from the antenna till its conversion to a digital format, is illustrated in the figure below. Based on data from the technical analysis, a VHF/UHF terrestrial interception system will receive a large variety of signals. Frequency ranges allocated for special applications cover an extended frequency band which overlaps certain frequency bands assigned to commercial applications. Modern communications systems with special applications use adaptive waveforms which can change the modulation scheme and the transfer rate so that the link quality will be maintained constant regardless of the changes encountered in the communications environment (interference, jamming). Extended frequency range available for certain special applications allows hiding used channels in guard bands or near radio channels assigned to commercial applications, making the detection of such signals more difficult. Modulation classes used primarily for VHF and UHF radio applications are CPFSK, PSK, and QAM. Most digital communication systems use synchronization sequences or training sequences which cause the occurrence of discrete spectral components in the signal power spectrum. These features can be exploited by signal detection algorithms. Narrowband digital radio channels analyzed use timing sequences positioned at the beginning of each broadcast or inserted in traffic. The use of periodic timing sequences determines the occurrence of discrete spectral components during the transmission of this sequence; these components constitute indications of active radio signals. For frequency hopping signals, timing sequences are inserted at the beginning of each hop. Detecting these sequences may offer information about the frequencies used during hopping and their succession in time. Commercial communications standards use training sequences, frequency correction or pilot carriers for synchronizing and controlling the power emitted. The structure of these sequences and the generated discrete spectral components were indicated for TETRA 1 and GSM signals. Chapter 3 examines the possibilities of detecting radio signals by energy measurements; this is one of the simplest detection methods of tactical radio signals. Radiometry is the most common form of implementation of the detection algorithm based on energy measurement. The performances of the digital radiometer for an AWGN channel type were characterized on hand of the detection probability and the false alarm probability. Its performances were evaluated for unmodulated signals, modulated signals, and narrowband digital waveforms. Due to advantages of digital signal processing, the digital radiometer shows superior performances compared to its analogical counterpart. The model of a digital baseband radiometer was represented by a MATLAB function. Based on the number of detections, the false alarm probability and the detection probability were estimated. Evaluation of resistance to interception of a VHF narrowband digital waveform was performed for the digital fixed frequency operation mode. The interception system used was the digital radiometer. Waveforms were simulated based on the technical parameters previously shown for this type of waveform. For the operating mode considered, the radio signal is of the 1REC MSK type with a 16 kbps symbol rate. In order to reduce the data processing time, the complex envelope corresponding to the simulated signal was also simulated. Based on the radiometer model described above the detection probability and the false alarm probability for the considered waveform were estimated. Results obtained were compared with theoretical values. According to the results obtained by simulation, it can be noticed that the detection probability increases with the SNR level. During simulation the threshold level and the number 4

5 of processed samples were kept constant. To investigate how the threshold level chosen influences the detection probability and the false alarm probability, simulations were repeated for different threshold level values, too. It can be noticed that, when choosing a threshold level close to the noise level, a high detection probability will be obtained even for small SNR values but it will remain constant even for great SNR values. Increasing the threshold level will generate lower values of the false alarm probability and a slower increase of the detection probability varying with SNR. Based on these observations, the use of a variable threshold level is recommended; this level should be lower for low SNR level and should increase with higher SNR. Reduced complexity allows the use of the radiometer for detection and interception of digital waveforms used in special applications. The digital version of the radiometer introduced in this paper has the advantage of flexibility; its parameters can be modified quickly. According to the results previously obtained one can draw the conclusion that the choice of parameters for implementing the radiometer has to consider the signal type and the accepted false alarm probability. The great diversity of waveforms used in special and commercial applications (narrowband/wideband/tdma/cdma/ofdm/fdma, frequency hopping, etc.) makes it very difficult to create a universal detection and interception system of these waveforms. The previously indicated simulations indicate the need to optimize detection algorithms based on signal type. Critical factors for the efficient operation of the radiometer are the correct estimation of the frequency band, the signal duration, and the noise level. The efficiency of detection methods can be increased by combining spectral estimation methods and time-frequency methods with the radiometer technique. Using spectral estimations allows the accurate delimitation of the occupied frequency bands; afterwards, by employing radiometer techniques, detection on each channel will be performed. This proves that the radiometer technique can be used in wideband systems. Simulations performed highlighted the convergence of the results obtained by simulation with the ones obtained using the analytical model when the channel considered is of the AWGN type. At the same time there may be differences between the theoretical model and the simulation when the modulation used creates variations in the modulated signal amplitude, thus causing variations of the useful signal. Chapter 4 presents contributions on detecting radio signals by power measurements. Based on the digital radiometer models developed (chapter 3), the possibility of detecting digital narrowband radio signals (used in special applications) and TETRA and GSM signals was tested. A digital spectral analyzer was used for the practical implementation of the radiometer. The measurement results are additionally processed in order to establish the presence or the absence of radio signals. The objective pursued is the optimization of the power measurement system parameters and of the processing algorithms used. Optimization was performed so that the detection probability would be maximized and the false alarm probability would be minimized. To assess these performance indicators it is necessary to estimate the probability function for the noise power and the useful signal power. Estimation of these features was performed based on the power samples measured with the PARZEN method. Different types of digital radio signals (TETRA, GSM) and different spectral analyzer parameters were used. Measurements demonstrated that for each signal type there is a set of optimum parameters of the measuring system. At the same time, using additional processing algorithms allows improving performances in comparison with the digital radiometer. Measurement errors of radio signals using the spectral analyzer depend on the measured signal characteristics and the parameters of the analyzer. Understanding how the measurement error is related to signal parameters and analyzer parameters allows estimating the error and correcting the final results. 5

6 The RF route of the spectral analyzer can be linearly or non-linearly, depending on the signal level at the first mixer input. The signal level at the mixer input has to be maintained below a certain threshold value so that it can operate linearly. Choosing a too small value will determine a decrease of the dynamic range available for measurements. Graphical methods used for establishing the optimum level of tactical signals at the mixer input were also indicated. The established value is optimum for imposed error and uncertainty values and is varies with the signal type and the spectral analyzer type used. When determining the optimum level, the TOI parameter of the spectral analyzer is used. This parameter is usually calculated by the producers for fixed values of the amplitude and spacing between 2 sinusoidal signals. In fact, the signals measured show different amplitude values for their spectral components as well as different spacing, so that the TOI parameter specified in the spectral analyzer catalogue does not always accurately describe the non-linear behavior of the mixer for the measured signals. The RF route contains a number of filters which can generate measurement errors when the signal measured occupies a band comparable to the one of the filters on the RF route. The band used by these filters is usually higher (> 2 MHz). The ratio of the measured signal band to the spectral analyzer resolution filter band causes the measurement of a smaller power than the real power at the resolution filter output. This error can be determined by simulation and corrected by adding it to the measured value. According to the simulations performed, this error can be neglected when the ratio of the measured signal band to the resolution filter band is smaller than 1. The error values computed by simulation are valid for CPFSK signals with constant envelope. Distribution of the measured power depends on the resolution filter band and the characteristics of the measured signals. Dispersion of power measured for GSM traffic channels (TCH) has high values because these channels do not have all time slots occupied. Estimated probability functions can be used for evaluate the detection probability and the false alarm probability. The analyses carried out can be completed with estimations of the probability function and for other signal types (frequency hopping, CDMA, etc.). It could also be useful to repeat these analyses for other radio channels, too. For estimating the probability function, other estimation methods could be employed. Chapter 5 examines the possibility of using time-frequency transforms (Gabor transform) for detecting frequency hopping or burst signals. A scenario used for detection and classification has to comprise the following processing operations: acquiring a frequency band at least equal with the hopping band; applying the Gabor transform on the discrete signal; spectrogram representation; eliminating stationary signals; detecting frequency hopping signals; estimating the hopping duration and the tuning duration. After detecting active frequencies, a representation in the time domain of the signal power at the respective frequency based on Gabor transform coefficients can be obtained. Power variation can be determined based on statistic indicators so that it can be decided whether the signal analyzed is stationary or not. For non-stationary signals, the emission duration and the tuning duration can be determined based on the power vs. time function. For estimating these parameters as accurately as possible, the effective duration of the Gaussian window has to be as small as possible. In order to obtain satisfying results, the effective duration value has to be at least half the tuning duration. Approximation errors of the signal envelope occur when there are great variations in the original signal envelope. Since the frequency resolution is equal with the inverted time resolution, the effective duration of the Gaussian window can be reduced however much in order to obtain a good spacing of the frequency hops. This is why the frequency resolution can be equal with the 6

7 frequency hopping channel bandwidth. For tactical radio stations with 25 khz spacing, time resolutions of 0.04 ms can be obtained. Taking into account that the tuning duration for this type of equipment is 1 ms, the resolution obtained is reasonably good. Chapter 6 investigates the detection possibilities by exploiting the characteristics of TETRA and GSM downlink signals. Timing and correction sequences were identified for signals used by the GSM system. The FCCH type slot can be used to improve the performances of detection algorithms. The length of the truncated duration of the baseband Gaussian impulse determines the instantaneous frequency during the FCCH slot duration. It was demonstrated that a value grater than 4 Ts (Ts duration of a symbol) determines negligible values of the frequency error. By simulation and by experiment it was marked out that transmitting a FCCH slot determines the occurrence of discrete spectral components at 67.7 khz spacing according to the type of correction sequence transmitted. At the same time, the instantaneous frequency has to assume the previously sent value during the transmission of the FCCH slot. These characteristics of FCCH type slots can be exploited in order to increase the performance of GSM signal detection algorithms. The instantaneous frequency analysis allows detecting BCCH channels. The value of the instantaneous frequency variance is used as an indicator for detecting noise and FCCH and TCH slots. The drawback of this algorithm is its impossibility to simultaneously process more than one channel, thus reducing the processing speed. At the same time, applying the algorithm implies computing the instantaneous frequency, so that its complexity is greater than the one of the detection algorithm by energy calculation or by correlation. The possibility of detecting GSM radio channels by correlation was experimentally proven. At the receiver, a waveform corresponding to the FCCH slot was used as a reference signal. Based on the value obtained at the correlation system output one can decide whether the processed signal corresponds to noise or to a FCCH/TCH slot. 3. MAIN CONTRIBUTIONS A. Experimental determination of the main characteristics of signals used in special applications (analogical FM, narrowband digital VHF, frequency hopping VHF); B. Experimental determination of the main characteristics of TETRA and GSM downlink signals; C. Estimating (by simulation) the performances of the digital radiometer for signal detection; D. Optimizing the spectral analyzer parameters for minimizing power measurement errors; E. Estimating the power probability function for measured power at the analyzer output with the PARZEN method for noise, GSM downlink signals and TETRA downlink signals; F. Establishing (by simulation) the optimum parameters of the Gabor transform for detecting FH signals; G. Experimental estimation of the detection capacity of GSM downlink signals by measuring the instantaneous frequency; H. Estimating the detection capacity of GSM downlink signals by correlation. 7