Detection and direction-inding o spread spectrum signals using correlation and narrowband intererence rejection Ulrika Ahnström,2,JohanFalk,3, Peter Händel,3, Maria Wikström Department o Electronic Warare Systems, Swedish Deence Research Agency Linköping, Sweden 2 E-mail: ulrika.ahnstrom@oi.se 3 KTH Signals, Sensors and Systems, Royal Institute o Technology Stockholm, Sweden Abstract An algorithm or correlation-based detection o direct sequence spread spectrum signals with direction inding, including direction-iltering and narrow-band intererence rejection, is implemented and evaluated in MATLAB. An analog noise-ree signal is generated and sampled by a test-bed system. Numerical simulations are run based on data corrupted by mutually uncorrelated white Gaussian noise sequences, and also with recorded noise rom two spatially separated HF radio receivers. The simulations and measurements show promising results or detection and direction-inding o covert wideband signals in low SNR and in presence o narrowband intererers. Direction iltering is shown to improve the results. Keywords detection, correlation, TDOA, DSSS, direction inding, intererence rejection, HF I. Introduction Direct sequence spread spectrum (DSSS) signals oer low probability o detection [] and protection against narrowband intererers. These properties have led to an increased use o DSSS signals in military communication applications during recent years. Thereore, methods or detection o signal presence and direction-inding o covert wideband signals with unknown characteristics are important components in an electronic warare system. The presence o narrowband intererers, as in the HF-band at -3 MHz, complicates the detection o covert signals. The considered method includes narrowband intererence rejection and requires no knowledge o the signal characteristics or detection and/or direction inding, such as bandwidth, modulation or spreading code. The considered correlation-based method or detection and direction-inding uses the outputs o two spatially separated receivers. The cross-correlation unction (CCF) and the cross-spectral density (CSD) are estimated rom the two received sequences. Detection is perormed in the requency domain by analysis o the phase and amplitude o the CSD. Timedierence o arrival (TDOA) based direction-inding Transmitter Intercept Intercept Intended Detection and TDOA estimation Fig.. A transmission system, including a transmitter and an intended receiver. Two intercept receivers are used by an electronic warare system or detection and/or positioning o the transmitter. is perormed in requency domain by estimating the phase-slope o the CSD [2]. Narrowband intererence rejection is implemented using digital notch ilters [3]. II. System overview The considered scenario is depicted in Figure, that is a military communication system consisting o a transmitter, an intended receiver and an electronic wararesystem. Thetransmissionsystemisassumed to use some measures (such as stealth waveorms, power control and directional antennas) to avoid detection and/or positioning by enemy orces. The electronic warare system uses two intercept receivers or detection and/or positioning and has no knowledge o waveorms or other signal characteristics. The transmitted signal s(t) is assumed to be a DSSS signal with unknown characteristics, such as bandwidth, modulation and spreading code. The signal is transmitted through a non-dispersive channel and is received by two synchronous spatially separated receivers. Sampling and quadrature mixing o the receiver outputs give the sequences x [k] and x 2 [k] which contain two noisy and dierent delayed versions o the transmitted signal. The sequences z [k] and z 2 [k] are used in order to describe antenna noise and internal receiver noise, that is
X [k] F compl. conj X *[] x [k] = s[k]+z [k] () x 2 [k] = s[k + ]+z 2 [k] (2) X 2 [k] F X 2 [] CSD where is the TDOA and s[k] is the sampled ( s = Hz) and quadrature mixed version o s(t). Theconsidered correlation based method or detection o signal presence and direction-inding uses the CCF and the CSD which are estimated rom the two received sequences x [k] and x 2 [k]. The CCF is deined as φ[m] =E[x [k + m] x 2[k]] (3) Since the involved signals are uncorrelated and s(t) is broadband, it ollows rom () and (2) that φ[m] =φ s (τ) τ=m+ (4) where φ s (τ) is the autocorrelation unction o s(t). TheCSDisthediscreteFouriertransormotheCCF, Φ[n] =F{φ[m]} = e j2π n/n Φ s [n] (5) The phase o Φ[n] reads Γ[n] =6 Φ[n] = 2πn N (6) The CSD is estimated rom {x [],.., x [N ],x 2 [],.., x 2 [N ]} as bφ[n] =F{x [k]} F {x 2 [k]} (7) A straight line Γ[n] b =bαn is itted to the phase slope 6 Φ[n] b within the signal bandwidth by the method o least squares. The linear phase o (6) is used to estimate rom b = bαn (8) 2π This approach is (almost) statistically eicient, in the sense that its error variance coincides with the Cramér-Rao bound [4] or SNRs above the threshold that is present in nonlinear estimation problems. Detection is perormed in requency domain by analysis o the phase and amplitude o the CSD. For covert low SNR signals, the CSD is oten to noisy to yield the inormation needed or detection and direction-inding. Direction-iltering by windowing the CCF improves the SNR in the CSD [5]. This is an eective method or broadband signals like those considered here. Applying the Fourier transorm on the windowed CCF gives the direction-iltered CSD (DFCSD) which is used or the detection and direction inding presented in this paper. A block diagram isgiveninfigure2. I CCF Direction iltering DFCCF DFCSD Fig. 2. The CSD is estimated rom the two received sequences, x [k] and x 2 [k]. Direction iltering by windowing the CCF improves the SNR in the CSD and gives the direction iltered cross spectral density DFCSD which is used or detection and direction inding. III. Intererence rejection The essence in the considered algorithm or detection and direction-inding is that the same signal is received by two spatially separated antennas with uncorrelated noise. This leads to a high correlation between the received sequences. When the received sequences contain narrowband intererers they reslut in a high correlation, similar as or the considered DSSS signal. This leads to a noisy DFCSD, and thus the alse-detection ratio increases as well as the variance in the direction-estimation. Narrowband intererence rejection is implemented using notch ilters [3]. In this paper a requency domain digital notch ilter is implemented to supress all high power (narrowband) signals. That is, all requency bins with an amplitude above a threshold are supressed. As the number o interering transmitters depends on the current requency band the threshold depends on the current CSD estimate. I the number o intererers is large there is a risk that some part o the DSSS signal is supressed as well. Thereore, the treshold must be chosen as a balance between intererence rejection and DSSS signal power. Simulations o typical electronic warare scenarios have shown satisactory results or a threshold that supress 5% o the requency bins. These are the requency bins with the highest amplitude in the received spectrum. Note that this threshold is adjusted to the current requency band. Figures 3-4 show the eects o intererence rejection on the CCF or a DSSS signal. In this example is strongly exaggerated, that is =, to separate the correlation caused by the DSSS signal. In Figure 3, the power spectral density (PSD) and the CCF or the received sequences are shown. It is impossible to distinguish the correlation caused by the DSSS signal. Figure 4 shows the PSD and CCF when the implemented intererence rejection is applied to the received sequences. The threshold or signal suppression is shown as a line in the PSD-plot. In this case the correlation caused by the DSSS signal is clearly separated rom the rest. One may note the dierent scales on the correlation axis.
CCF Relative power [db] CCF Relative power [db] abs(dfcsd) -2 3 db -4-6 -8 - -2-25 -2-5 - -5 5 5 2 25 Frequency [khz] 2 Frequency x 9 phase(dfcsd) 8 6 4 2-6 -4-2 2 4 6 Delay [Number o samples] x 4 Fig. 3. Two dierently delayed versions, =, o a DSSS signal added to noise sequences recorded rom the HF-band gives the PSD and CCF plotted above. It is impossible to distinguish the correlation caused by the DSSS signal. 2 Frequency Fig. 5. Detection is done in two steps. First, the bandwidth o the DSSS signal is estimated by amplitude detection. Astraightlineisitted to the phase within the estimated bandwidth and the phase curve is used to decide whether asignalispresentornot. -2-4 -6-8 - -2-25 -2-5 - -5 5 5 2 25 Frequency [khz].5.5 2 x 7 iltering and narrowband intererence rejection, is implemented and evaluated in MATLAB. An analog noise-ree DSSS signal is generated and sampled by an experimental system, developed by the Swedish Deence Research Agency [6]. The experimental system consists o a programmable waveorm generator and a commercial receiver with digital tuner, down converter and A/D-converter, see Table. The generated DSSS signal is similar to a 3G mobile network signal [7]. The bandwidth is 3%. Moreover, the method is also tested on a military stealth signal denoted. -6-4 -2 2 4 6 Delay [Number o samples] x 4 Fig. 4. Ater applying the implemented intererence rejection the correlation caused by the DSSS signal is clearly separated rom the rest. The threshold or signal suppression is shownasalineinthepsd. IV. Phase-analysis based detection Detection is traditionally done by threshold detection within the bandwidth o the DFCSD [5]. Here a method or detection o signal presence is presented that besides amplitude analysis uses the phaselinearity. The bandwidth o the DSSS signal is estimated by amplitude detection o the CSD estimated rom (7), see Figure 5. A straight line is itted to the phase slope within the estimated bandwidth and b is estimated rom (8) [2], [4]. The mean square error between the straight line and the phase curve is used to decide whether a signal is present or not. V. Numerical results The presented algorithm or correlation-based detection with direction-inding, including direction- TABLE I Technical speciications o experimental system Waveorm generator Signal generator Frequency range (receiver) A/D converter Sampel requency Decimation Sample requency ater decimation Eective bandwidth AMIQ (Rhode&Schwarz) SMHU58 (Rhode&Schwarz) SCR5-B (Andrew SciComm) 2-34 MHz 2 bit 28.5 MHz 4times 7.25 MHz 5.7 MHz Numerical simulations are run based on data corrupted by mutually uncorrelated white Gaussian noise sequences, and also with recorded noise rom two spatially separated HF-receivers. Two dierently delayed versions o the signal s[n] represent the signal rom the spatially separated receivers. Sequences representing antenna noise and internal receiver noise are added to the signal parts as shown in Figure 6.
Probability o detection MSE [db] Probability o detection noise channel signal present & ^ s[n] s [n] = s[n] Detection & Direction-inding.8 s 2 [n] = s[n- ].6 noise channel 2 signal not present.4 Fig. 6. Noise sequences are added to two dierently delayed versions o the signal, s[n]..2 The signal-to-noise ratio is deined as SNR = log( E s ) E n where E s is the signal power and E n the noise power within signal bandwidth. When the noise contains narrowband intererers the signal-to-intererence ratio (SIR) is deined as SIR = log( E n E s bw s ) bw tot where bw s is the signal bandwidth and bw tot the ull received bandwidth. First simulations are run with white Gaussian noise sequences. The probability o detection based on Monte Carlo simulations is estimated or dierent SNRs. The mean square error o the TDOA-estimate is calculated. As the probability o detection is related to the alse-detection ratio, simulations are run to estimate that as well. Simulations are also run with recorded noise rom two HF-receivers separated by a distance o 75 m. The sample rate is 578 Hz and the center requency 9 MHz. The upper plot in Figure 3 shows a typical spectrum or the received noise sequences. The probability o detection based on 2 dierent recorded noise sequences is estimated or dierent SIRs. The mean square error o b is also calculated. The alsedetection ratio is estimated or the received noise sequences as well. The results o the numerical simulations are shown in Figures 7-. -3-25 -2-5 - -5 SNR [db] Fig. 7. The probability o detection or dierent SNRs, based on Monte Carlo simulations. The alse detection ratio is estimated to 5%. 2 - -2-3 -4-3 -25-2 -5 - -5 SNR [db] Fig. 8. The mean square error o the TDOA-estimate, b, or dierent SNR, based on Monte Carlo simulations...9.8.7.6.5 VI. Conclusions An algorithm or correlation-based detection o DSSS signals with direction inding, including direction iltering and narrowband intererence rejection is implemented and evaluated in MATLAB. Numerical simulations based on data corrupted by mutually uncorrelated white Gaussian noise sequences shows that the probability o detection is more than 6% or SNRs above -5 db..4.3.2-3 -25-2 -5 - -5 SIR [db] Fig. 9. The probability o detection or dierent SIRs, based on simulations with 2 dierent recorded noise sequences rom the HF-band. The alse detection ratio is estimated to 9%.
MSE [db] 5-5 - -5-2 -25-3 -35-4 -45-3 -25-2 -5 - -5 SIR [db] Fig.. The mean square error o the TDOA-estimate, b, or dierent SIR, based on simulations with 2 dierent recorded noise sequences rom the HF-band. Simulations run with recorded noise rom two separated HF-receivers show that narrowband intererence rejection is necessary when the received sequences contain narrowband intererers. Simulations including intererence rejection by a requency domain digital notch ilter show promising results or detection and direction inding o DSSS signals at the HF-band and in low SIR. Simulations also show that detection and direction inding is improved by direction iltering by windowing the CCF. Reerences [] A. M. Wik, A. L. Lindblad, Novel concept using iltered spreading codes, IEEE Military Communications Conerence MILCOM 96, McLean, VA, Oct. 2-24, 996. [2] A. G. Piersol, Time delay estimation using phase data, IEEE Transaction on Acoustics, Speech, Signal Processing, June 98, Vol. 29, No.3, pt. 2, pp. 47-477. [3] N. Hallqwist, B. Lagerquist, Intererence rejection techniques in a DS spread-spectrum HF radio system, Report FOA-R- -94-66-3.5- -SE, Deence Research Establishment, Sweden 994. [4] J. Falk, P. Händel, M. Jansson, Direction inding or electronic warare systems using the phase o the cross spectral density, Radiovetenskap och Kommunikation (RVK), Stockholm, Sweden 22. [5] A. W. Houghton, C. D. Reeve, Detection o spreadspectrum signals using the time-domain iltered cross spectral density, IEE Proceedings - Radar, Sonar and Navigation, December 995, Vol. 42, No 6, pp. 286-292. [6] P Johansson, Rasmus - A spread spectrum and modulation evaluation system, Report FOA-R- --439-54- - SE, Deence Research Establishment, Sweden 2. [7] 3GPP, Universal Mobile Telecommunications System (UMTS); Spreading and modulation (FDD), TS 25.23 v3.., ETSI, January 2