Low-Slow-Small (LSS) Target Detection Based on Micro Doppler Analysis in Forward Scattering Radar Geometry

The increase in drone misuse by civilian apart from military applications is alarming and need to be addressed. This drone is characterized as a low altitude, slow speed, and small radar cross-section (RCS) (LSS) target and is considered difficult to be detected and classified among other biological targets, such as insects and birds existing in the same surveillance volume. Although several attempts reported the successful drone detection on radio frequency-based (RF), thermal, acoustic, video imaging, and other non-technical methods, however, there are also many limitations. Thus, this paper investigated a micro-Doppler analysis from drone rotating blades for detection in a special Forward Scattering Radar (FSR) geometry. The paper leveraged the identified benefits of FSR mode over conventional radars, such as improved radar cross-section (RCS) value irrespective of radar absorbing material (RAM), direct signal perturbation, and high resolutions. To prove the concept, a received signal model for micro-Doppler analysis, a simulation work, and experimental validation are elaborated and explained in the paper. Two rotating blades aspect angle scenarios were considered, which are (i) when drone makes a turn, the blade cross-sectional area faces the receiver and (ii) when drone maneuvers normally, the cross-sectional blade faces up. The FSR system successfully detected a commercial drone and extracted the micro features of a rotating blade. It further verified the feasibility of using a parabolic dish antenna as a receiver in FSR geometry; this marked an appreciable achievement towards the FSR system performance, which in future could be implemented as either active or passive FSR system.


Introduction
The low cost, off the shelf availability, and operational flexibility of a drone lead to its misuse by civilian society, apart from military applications, thereby increasing posing threats to society [1]. Potential threats, such as drugs smuggling, transportation of contraband materials like weapons to criminals in prison, unauthorized imaging and filming in restricted areas, illegal surveillance, air collision, terrorist attacks, and radio frequency (RF) jamming are among the potential threats posed by the drone misuse. This has become a matter of importance due to the alarming threat to our community and the recorded security breach [2] caused by a drone. It may have posed a threat to scenarios, such as (i) our recreational centers like a stadium, (ii) strategic and critical infrastructure or secured environment, etc., such that no sound can be noticed and no warnings may be provided. may be caused by rotation, vibrations, or both [19,20], resulted from multiple combinations of point scatterers [21].
The FSR geometry as presented in Figure 1, had the target moving toward the baseline. The scattered signal caused an additional modulation due to the rotating blades, thus, forwarded to the receiver as received signal. From Figure 1, a combination of both the scattered and the direct signal forms the received signal Srx = Sdir + Ssc. Assuming the direct signal is given by Sdir = AdirCos(ωc)t, and the scattered signal be Ssc = AscSin(ωc + ωsc)t, the received signal, as in [22], is given by Srx = AdirCos(ωc)t + AscSin(ωc + ωsc)t (1) where ωc = 2πFc and ωsc = 2πFd. Equation (1) is the representation of the entire signal at the receiver. Within the scattered signal, a Doppler component is present. The received signal passes through a non-linear device and yields a signal with a square law on the output. A filtering process is then applied using a low pass filter (LPF) to filter out the high frequency component from the diode output, leaving only the two spectral components, i.e., the direct leakage signal and the signal containing the Doppler component scattered by the target and a conversion factor due to the type of diode used. The direct signal (leakage) component can be filtered out by using a high pass filter (HPF), leaving the signal with only a Doppler component, thus, ωsc is the angular frequency combining the Doppler component fd due to linear motion and the micro-Doppler fmd due to the blade rotation; thus, ωsc = 2πfdt + 2πfmdt, as described in [22], and the received signal equation is now: Srx(t) = Adir * Asc * Sin (2πfd + 2πfmd)t (2) From [22], the general Doppler equation for the bistatic arrangement is given by where V is the radial velocity, β is the Tx-target-Rx bistatic angle, [cos ( )cos ()] are the coefficient factors due to bistatic geometry, and λ is the wavelength. A detailed derivation of the Doppler equation is also in [22]. In this paper, an actual quadcopter drone geometry with 4-rotors of 2-rotating blades, maintaining similar dimensions was represented by a crossbar in FSR geometry, as illustrated in Figure 2. In the geometry, the drone was placed equidistance between the transmitter and the receiver. Two significant scenarios, such as horizontally (Facing-Up) and vertically (Facing-Rx) aligned, 1 m away from the baseline, were modeled. If the Tx is located at the origin of fixed coordinates (X0, Y0, Z0 = 0) and the Rx placed on the same plane of equal height separated by a baseline distance BL, the corresponding Rx coordinate is given by (BL, 0, 0). The rotors positions were assumed to be at the tip ends of the crossbar, and the rotor vectors signifying their positions on the (xyz) plane was defined. In the 'Facing-Rx' scenario, the blades of all rotors were made to rotate about the Z-axis, and the X-axis for 'Facing-Up' scenarios, as illustrated in Figure 2a  From Figure 1, a combination of both the scattered and the direct signal forms the received signal S rx = S dir + S sc . Assuming the direct signal is given by S dir = A dir Cos(ω c )t, and the scattered signal be S sc = A sc Sin(ω c + ω sc )t, the received signal, as in [22], is given by S rx = A dir Cos(ω c )t + A sc Sin(ω c + ω sc )t (1) where ω c = 2πF c and ω sc = 2πF d . Equation (1) is the representation of the entire signal at the receiver. Within the scattered signal, a Doppler component is present. The received signal passes through a non-linear device and yields a signal with a square law on the output. A filtering process is then applied using a low pass filter (LPF) to filter out the high frequency component from the diode output, leaving only the two spectral components, i.e., the direct leakage signal and the signal containing the Doppler component scattered by the target and a conversion factor due to the type of diode used. The direct signal (leakage) component can be filtered out by using a high pass filter (HPF), leaving the signal with only a Doppler component, thus, ω sc is the angular frequency combining the Doppler component f d due to linear motion and the micro-Doppler f md due to the blade rotation; thus, ω sc = 2πf d t + 2πf md t, as described in [22], and the received signal equation is now: From [22], the general Doppler equation for the bistatic arrangement is given by where V is the radial velocity, β is the Tx-target-Rx bistatic angle, [cos ( β 2 ) cos(δ)] are the coefficient factors due to bistatic geometry, and λ is the wavelength. A detailed derivation of the Doppler equation is also in [22].
In this paper, an actual quadcopter drone geometry with 4-rotors of 2-rotating blades, maintaining similar dimensions was represented by a crossbar in FSR geometry, as illustrated in Figure 2. In the geometry, the drone was placed equidistance between the transmitter and the receiver. Two significant scenarios, such as horizontally (Facing-Up) and vertically (Facing-Rx) aligned, 1 m away from the baseline, were modeled. If the T x is located at the origin of fixed coordinates (X 0 , Y 0 , Z 0 = 0) and the R x placed on the same plane of equal height separated by a baseline distance B L , the corresponding R x coordinate is given by (B L , 0, 0). The rotors positions were assumed to be at the tip ends of the crossbar, Sensors 2019, 19, 3332 4 of 20 and the rotor vectors signifying their positions on the (xyz) plane was defined. In the 'Facing-Rx' scenario, the blades of all rotors were made to rotate about the Z-axis, and the X-axis for 'Facing-Up' scenarios, as illustrated in Figure 2a,b. The corresponding position vectors of all the rotors for each case are described in Equations (4) and (5). If the rotor vectors are represented by Rna and Rnb, (for n = 1… 4), where n corresponds to the number of rotors, 'a' and 'b' are the corresponding scenarios for 'Facing-Up' and 'Facing-Rx', respectively, D is the distance travel, and Wd is the width of the drone, the corresponding rotor vectors are represented by: From Equations (4) and (5), R1a represents rotor-1 of Figure 2a and R1b refers to rotor-1 of Figure  2b. Each of this rotor serves as the center of 2-blades upon which the blades rotate. Based on the rotor positions of Figure 2, a receiver-range geometry of two rotating blades and their corresponding range If the rotor vectors are represented by R na and R nb , (for n = 1 . . . 4), where n corresponds to the number of rotors, 'a' and 'b' are the corresponding scenarios for 'Facing-Up' and 'Facing-Rx', respectively, D is the distance travel, and W d is the width of the drone, the corresponding rotor vectors are represented by: From Equations (4) and (5), R 1a represents rotor-1 of Figure 2a and R 1b refers to rotor-1 of Figure 2b. Each of this rotor serves as the center of 2-blades upon which the blades rotate. Based on the rotor positions of Figure 2, a receiver-range geometry of two rotating blades and their corresponding range profiles are presented in Figure 3. The ranges between the receiver-to-blade-tip are given by R rx _tip nm , and the transmitter-blade-tip by R tx _tip nm . These make a bistatic angle 'β tip ' for each rotor 'n' blade 'm'. Similarly, R tx _Fixed n and R rx _Fixed n are the corresponding ranges for transmitter-to-rotor fixed position and rotor-to-receiver, with bistatic angle 'β'. The scattered signal can, therefore, be calculated based on the location and orientation of the blades. and the transmitter-blade-tip by Rtx_tipnm. These make a bistatic angle 'βtip' for each rotor 'n' blade 'm'. Similarly, Rtx_Fixedn and Rrx_Fixedn are the corresponding ranges for transmitter-to-rotor fixed position and rotor-to-receiver, with bistatic angle 'β'. The scattered signal can, therefore, be calculated based on the location and orientation of the blades.  Figure 3 is a zoomed view of only one rotor described in Figure 2. Each rotor has 2-blades rotating about a fixed rotor point that forms the bistatic angle between the receiver and the transmitter. It describes the Doppler due to the blade-tip and the blade's range profile referenced to both transmitter and the receiver. The 'βtip' is the bistatic angle formed by the blade-tip while rotating through 0-2π, about a fixed rotor position. The variation of the blade yaw angle is determined by the rotation rate, the blade length Lb, and the initial angle given by 'θ'. If the blade rotates at a constant rotation rate (Ω) over time (t), a yaw angle is formed (Ωt + θo), and, thus, the velocity is caused by the blade tip. The velocity of the blade-tip (Vtip), which is a function of the Doppler generated, is indicated in (6).
where Lb is the blade length, 2π is the conversion rate of radial velocity. The received range reference to the blade tips can be obtained by using (7) Rrx_tipnm = [Rrx_Fixedn 2 +Lb 2 − (2 * Rrx_Fixedn * Lb) * Cos(θ + θn) ] 1/2 where Rtx_Fixedn and Rrx_Fixedn are the corresponding transmitter and receiver ranges for the rotor 'n'. These ranges are used to obtain the bistatic angle 'β' for the target's radial motion. Similarly, Rrx_tipnm and Rtx_tipnm are the corresponding ranges of receiver-to-blade-tip and transmitter-to-bladetip, for rotor 'n', blade 'm'. These are also used to calculate the bistatic angle 'βtip' made by the bladetip at each point while rotating. From (2), the total received signal for rotor 'n' with 'N' maximum number of blades, the sum of the received signal is, therefore, represented as:  Figure 3 is a zoomed view of only one rotor described in Figure 2. Each rotor has 2-blades rotating about a fixed rotor point that forms the bistatic angle between the receiver and the transmitter. It describes the Doppler due to the blade-tip and the blade's range profile referenced to both transmitter and the receiver. The 'β tip ' is the bistatic angle formed by the blade-tip while rotating through 0-2π, about a fixed rotor position. The variation of the blade yaw angle is determined by the rotation rate, the blade length L b , and the initial angle given by 'θ'. If the blade rotates at a constant rotation rate (Ω) over time (t), a yaw angle is formed (Ωt + θ o ), and, thus, the velocity is caused by the blade tip. The velocity of the blade-tip (V tip ), which is a function of the Doppler generated, is indicated in (6).
where L b is the blade length, 2π is the conversion rate of radial velocity. The received range reference to the blade tips can be obtained by using (7) R rx _tip nm = [R rx _Fixed n 2 + L b 2 − (2 * R rx _Fixed n * L b ) * Cos(θ + θ n )] 1/2 where R tx_ Fixed n and R rx_ Fixed n are the corresponding transmitter and receiver ranges for the rotor 'n'. These ranges are used to obtain the bistatic angle 'β' for the target's radial motion. Similarly, R rx _tip nm and R tx _tip nm are the corresponding ranges of receiver-to-blade-tip and transmitter-to-blade-tip, for rotor 'n', blade 'm'. These are also used to calculate the bistatic angle 'β tip ' made by the blade-tip at each point while rotating. From (2), the total received signal for rotor 'n' with 'N' maximum number of blades, the sum of the received signal is, therefore, represented as: Φ md (t) is the phase modulation generated as a result of a circumferential micromotion of each blade and can be represented as L b is the blade length, Ωt and θ o are the rotation rates of the rotating blade in the rs−1 and initial angle of the blade, respectively. If a rotor has multiple numbers of blades, the rotating scatterers of each blade with different initial angles [23] can be generated by using θ n = θ o + n2π/N, for n = 0, 1, . . . , N − 1; in our case, the rotor had only 2-blades, therefore, n = 0, 1.
As shown in Figure 3, the angle 'θ o ' was assumed to lie at the origin, and this made the initial angle 'θ o ' to be zero [18], leaving only the blade rotation rate Ωt. We used the Bessel function of the first kind to obtain the micro-Doppler frequency of each blade. This was achieved by differentiating Equation (9) for time 't' [24,25].
where V tip is the velocity of the blade tip. If 2π is the conversion rate of the angular speed Ω t , L b is the blade's length, then V btip = 2π * L b * Ω t . The maximum micro-Doppler is now represented as The total signal received for 'N' number of blades can, therefore, be represented as If the received amplitude A rx(n) is represented as A r * A e (n), then the total received signal is In summary, the received signal of each of the four rotors can be treated independently based on their geometrical positions.
Based on the described geometry of Figures 2 and 3 ' is the target radius. These frequencies were also used during the experimental micro-Doppler analysis. The corresponding short-time Fourier transform (STFT) GHz S-band (for air traffic control), 5 GHz C-band (for weather radars), in addition to 7 GHz, were used to demonstrate the simulated micro-Doppler of the rotating blades. Each frequency used satisfied the equation 2πƖ/λ >10 in a way that the target dimension was electrically large when compared with the wavelength [26], where 'Ɩ' is the target radius. These frequencies were also used during the experimental micro-Doppler analysis. The corresponding short-time Fourier transform (STFT) signature representation is illustrated in Figure 4. Figure 4a-c resulted from Facing-Rx geometry of Figure 2b, while Figure 4c,d is the signature due to Facing-Up scenario of Figure 2a. A micro-Doppler due to the rotation of each blade-tip, while the rotor was in 'On' position, for an observation time of 1 s, is presented in Figure 4, with both blades preceding one another. The signal response of each blade was generated independently, based on the geometrical position to their corresponding rotor position. The signature of each blade was represented as R 11 , R 12 , R 31 , R 32 with only rotor R 2 and R 4 mapped into one another. It was observed that the Facing-Up scenario yielded better resolution by separating the two mapped rotors R 2 and R 4 , as can be seen in Figure 4d-e. Result in Figure 4 shows the resolution improved by increasing the operating frequency. This is because micro-Doppler is proportional to the operating frequency. This revealed how smaller wavelength is expected to yield a better Doppler resolution during the experimental work. Based on the simulated results, the signatures due to rotating blades corresponds to the findings in [17,19]. The observation angle played a vital role in the Doppler and range resolutions, as can be seen between 'Facing-Up' and 'Facing-Rx' blade orientation. In the 'Facing-Up' scenario of Figure 4d-f, the overlapping rotors R 2 and R 4 can be seen, unlike the 'Facing-Rx' case.

Experimental Setup
The experiments conducted in this paper consisted of two parts, as described in Figure 5: i.
The first part aimed to validate the micro-Doppler theoretical model, as described in Section 2. The first part comprised of an experiment in the anechoic chamber and outdoor experiment with the low elevation of the transmitter and FSR receiver. ii.
The second part aimed to study the feasibility of the proposed FSR system to detect the drone's linear motion and to extract the micro-Doppler of the rotating blade.  The first part comprised of an experiment in the anechoic chamber and outdoor experiment with the low elevation of the transmitter and FSR receiver. ii.
The second part aimed to study the feasibility of the proposed FSR system to detect the drone's linear motion and to extract the micro-Doppler of the rotating blade. Figure 5 describes the general composition of the experiments conducted in this study.
For the validation, the experiment was conducted with fixed rotating blades for outdoors and in the anechoic chamber.

Signal Processing and Receiver Hardware
The FSR has simple receiver circuits that do not require any synchronization channel and, thus, contains only one receiving channel with a self-mixing technique using a non-linear amplitude detector (Diode). The signal arriving the receiving antenna is composed of a superposition of two signals, the direct path signal from the transmitter and the echoed signal by the target. The signal composition is directly fed into the low noise amplifier (LNA). The amplitude detector performs a demodulation process on the amplified signal from the LNA output and then extracts a very lowfrequency band from the phase-amplitude modulated signal. The low pass filter (LPF) outputs the very low Doppler signal content and discards the high-frequency component. This signal was saved on the PC for further processing. Figure 6 describes the hardware composition of the FSR receiver circuit. During the experiment, the signal sampling frequency of fs = 50 kHz knows the expected Doppler due to the rotating blades. This was further down-scaled by 10 to fs = 5 kHz for unambiguous

Signal Processing and Receiver Hardware
The FSR has simple receiver circuits that do not require any synchronization channel and, thus, contains only one receiving channel with a self-mixing technique using a non-linear amplitude detector (Diode). The signal arriving the receiving antenna is composed of a superposition of two signals, the direct path signal from the transmitter and the echoed signal by the target. The signal composition is directly fed into the low noise amplifier (LNA). The amplitude detector performs a demodulation process on the amplified signal from the LNA output and then extracts a very low-frequency band from the phase-amplitude modulated signal. The low pass filter (LPF) outputs the very low Doppler signal content and discards the high-frequency component. This signal was saved on the PC for further processing. Figure 6 describes the hardware composition of the FSR receiver circuit. During the experiment, the signal sampling frequency of f s = 50 kHz knows the expected Doppler due to the rotating blades. This was further down-scaled by 10 to f s = 5 kHz for unambiguous generated Doppler during the offline Matlab signal processing. For the rotating blade signature, a clear description between the 'On' position of the rotating blade, the 'Off' state, and their corresponding transition states is presented in all cases. For drone detection, the same signal processing was maintained after the signal was received.

Experimental Results
In this study, Sections 4.1 and 4.2 focus on the "micro-Doppler" components due to the rotating blades and are compared with the simulation. The experiment in Section 4.3 considered both the main Doppler due to drone linear motion and the micro-Doppler due to the blade micro motions. The experiment is more practical due to the baseline range, location of the transmitter, and antenna specifications. For Sections 4.1 and 4.2, the experiment used "horn antenna" at both transmitting and receiving end, while in Section 4.3, the 11 GHz experimental setup involved a horn antenna transmitter and a parabolic dish receiver. In Section 4.3, we considered using 11 GHz due to the following reasons: a) To have a more meaningful and practical FSR system, future work will involve passive system by using illuminator of opportunity. For drone detection using FSR geometry, the best

Experimental Results
In this study, Sections 4.1 and 4.2 focus on the "micro-Doppler" components due to the rotating blades and are compared with the simulation. The experiment in Section 4.3 considered both the main Doppler due to drone linear motion and the micro-Doppler due to the blade micro motions. The experiment is more practical due to the baseline range, location of the transmitter, and antenna specifications. For Sections 4.1 and 4.2, the experiment used "horn antenna" at both transmitting and Sensors 2019, 19, 3332 9 of 20 receiving end, while in Section 4.3, the 11 GHz experimental setup involved a horn antenna transmitter and a parabolic dish receiver. In Section 4.3, we considered using 11 GHz due to the following reasons: (a) To have a more meaningful and practical FSR system, future work will involve passive system by using illuminator of opportunity. For drone detection using FSR geometry, the best illuminator is using a satellite signal; thus, the proposed 11 GHz in our study could resemble satellite signal working at K/Ka/Ku band. (b) Based on the satellite signal as mentioned in (a), the system needs to use a dish antenna for acceptable Signal to Noise (SNR). Thus, practically using lower frequency will make the dish antenna bigger, so at 11 GHz, the antenna size is reasonable. (c) To verify the performance of our developed receiving system that only operates within Ku-band frequencies, in the detection of a real commercial drone.
The experiment aimed at detecting and extracting the micro-Doppler of rotating blades. The blade is 0.12 m length, which made to rotate at an approximate speed of 36 r/s. Firstly, the blades are wrapped with a conducting material (usually aluminum foil) to improve its reflective property, because most drones are made of either plastic material or carbon fiber [27]. Subsequently, the default blades are used without been wrapped to determine the degree of their detectability as compared to the first case. During the experiment, scenarios involved in both cases include blades Facing-Receiver, Facing-Up, and maneuvered motion. The effect of look angles of the drone blade on the rotating blade while detection in addition to wavelength selection was also considered. This paper compared the simulated and the experimental signature for 3 GHz operating frequency, as illustrated in Figure 7.

Rotating Blades Outdoor
An outdoor experiment was conducted to ascertain the detectability of the drone rotating blades. It further verified the look angle effects and the effects due to the maneuvered motion of the drone. To achieve these, we used 3 GHz frequency for both Facing-Receiver and Facing-Up scenarios for illustration purposes. This was based on the observed similarity in the signature for all the three  The three separated bars represented the signatures of R 1 , R 2 -R 4 (R 2 and R 4 mapped on one another), and R 3 , respectively. The 'start' and 'stop' states represented the blade's transition when switched 'On' and 'Off', respectively. In both cases, blade-1 and blade-2 mapped together. Similarly, the simulation represented only the 'On' position of the rotating blade and had R 11 , R 12 , R 31 , R 32 , represented with R 2 and R 4 mapped together, as described earlier. Besides, unlike the experimental signature that represented the signature of the entire blade length, the simulated signature only represented the blade-tips. The variation in the Doppler value resulted due to multiple factors, such as the difference in obtaining the exact value of the theoretical parameters like the bistatic angle, aspect angle, blade rotation rate, etc., and the signal degradation due to other external factors.

Rotating Blades Outdoor
An outdoor experiment was conducted to ascertain the detectability of the drone rotating blades. It further verified the look angle effects and the effects due to the maneuvered motion of the drone. To achieve these, we used 3 GHz frequency for both Facing-Receiver and Facing-Up scenarios for illustration purposes. This was based on the observed similarity in the signature for all the three operating frequencies earlier used. Subsequently, the experimental setup, shown in Figure 8, was not to scale but rather for illustration purposes to provide insightful information on how the experiment was conducted.
represented only the 'On' position of the rotating blade and had R11, R12, R31, R32, represented with R2 and R4 mapped together, as described earlier. Besides, unlike the experimental signature that represented the signature of the entire blade length, the simulated signature only represented the blade-tips. The variation in the Doppler value resulted due to multiple factors, such as the difference in obtaining the exact value of the theoretical parameters like the bistatic angle, aspect angle, blade rotation rate, etc., and the signal degradation due to other external factors.

Rotating Blades Outdoor
An outdoor experiment was conducted to ascertain the detectability of the drone rotating blades. It further verified the look angle effects and the effects due to the maneuvered motion of the drone. To achieve these, we used 3 GHz frequency for both Facing-Receiver and Facing-Up scenarios for illustration purposes. This was based on the observed similarity in the signature for all the three operating frequencies earlier used. Subsequently, the experimental setup, shown in Figure 8, was not to scale but rather for illustration purposes to provide insightful information on how the experiment was conducted.

Blade Detection
Total observation time of 15 s was used to verify the detectability of the rotating blade, which was considered enough to record the three-states transition of the blades, i.e., Off-On-Off. The blade was initially in 'Off' position, then switched 'On' and then 'Off' again for 5 s each. Between each 'On' and 'Off' positions, a transition time was observed with a projected rise due to starting torque from 'steady' state to 'rotation' state, as seen in Figure 9. A similar occurrence happened when the rotors slowed down to steady state after it was turned 'Off'. Both time domain and the STFT represented only 9 s because the transition time of "2 s" out of "5 s" for both start and stop were captured, while the remaining "5 s" was for "On" state. Despite the weather and other effects, such as ground reflection, clutter, and volume scattering, the detected signature was considered a success; hence, ascertained our prior claims and assumption in the designed theoretical model. Total observation time of 15 s was used to verify the detectability of the rotating blade, which was considered enough to record the three-states transition of the blades, i.e., Off-On-Off. The blade was initially in 'Off' position, then switched 'On' and then 'Off' again for 5 s each. Between each 'On' and 'Off' positions, a transition time was observed with a projected rise due to starting torque from 'steady' state to 'rotation' state, as seen in Figure 9. A similar occurrence happened when the rotors slowed down to steady state after it was turned 'Off'. Both time domain and the STFT represented only 9 s because the transition time of "2 s" out of "5 s" for both start and stop were captured, while the remaining "5 s" was for "On" state. Despite the weather and other effects, such as ground reflection, clutter, and volume scattering, the detected signature was considered a success; hence, ascertained our prior claims and assumption in the designed theoretical model.

Aspects of Angle Effect
The effect of aspect angle due to blades orientation was verified by using the two scenarios, i.e., the blades 'Facing-Rx' and the blades 'Facing-Up' orientations, as described in Figure 5. This was

Aspects of Angle Effect
The effect of aspect angle due to blades orientation was verified by using the two scenarios, i.e., the blades 'Facing-Rx' and the blades 'Facing-Up' orientations, as described in Figure 5. This was described by using 1-rotor, 2-blades and 2-rotors, 2-blades to display Doppler resolutions. Maintaining similar operating frequency of 3 GHz, the detected signatures are illustrated in Figure 10.

Aspects of Angle Effect
The effect of aspect angle due to blades orientation was verified by using the two scenarios, i.e., the blades 'Facing-Rx' and the blades 'Facing-Up' orientations, as described in Figure 5. This was described by using 1-rotor, 2-blades and 2-rotors, 2-blades to display Doppler resolutions. Maintaining similar operating frequency of 3 GHz, the detected signatures are illustrated in Figure  10. A short-time Fourier transform (STFT) signature representation of 1-rotor, 2-blades and 2-rotors, 2-blades, with the rotating blades facing the receiver is shown in Figure 10a,b. Similarly, with the rotating blades Facing-Up, their signatures can also be seen in Figure 10c,d. Two bars represented the 2-rotors signature (R1 and R3) during 'On' position, with blade 1 and 2 mapped on one another. It was observed that the 'Facing-Rx' orientation had higher power intensity as compared to the 'Facing-Up' scenario. In both cases, good Doppler resolutions were observed as both rotors signature were seen. A short-time Fourier transform (STFT) signature representation of 1-rotor, 2-blades and 2-rotors, 2-blades, with the rotating blades facing the receiver is shown in Figure 10a,b. Similarly, with the rotating blades Facing-Up, their signatures can also be seen in Figure 10c,d. Two bars represented the 2-rotors signature (R 1 and R 3 ) during 'On' position, with blade 1 and 2 mapped on one another. It was observed that the 'Facing-Rx' orientation had higher power intensity as compared to the 'Facing-Up' scenario. In both cases, good Doppler resolutions were observed as both rotors signature were seen.

Maneuvering Effects
Based on the drone operational behavior, we considered the blade signature due to maneuvering motion. At first, the blade was made to be in 'On' state facing the receiver with blades rotating at an approximate speed of 36 r/s with 3 GHz operating frequency. The entire 4-rotors would then be turned slowly about a fixed position in a clockwise direction over a complete circle 0-2π. The observed change in a signature over 360 • was observed in both time domain and spectrogram representation, as shown in Figure 11. The first part of the time domain was when the blade was facing in the direction of the receiver, i.e., perpendicular to the baseline. The blades turned by 90 • to face down, then faced the transmitter, and finally faced up. The signal power seemed to always be of higher intensity when facing the receiver or transmitter but degraded when facing up or down. This signature might be of help in recognizing the drone maneuvering motion in the course of actual drone flight operation.
motion. At first, the blade was made to be in 'On' state facing the receiver with blades rotating at an approximate speed of 36 r/s with 3 GHz operating frequency. The entire 4-rotors would then be turned slowly about a fixed position in a clockwise direction over a complete circle 0-2π. The observed change in a signature over 360° was observed in both time domain and spectrogram representation, as shown in Figure 11. The first part of the time domain was when the blade was facing in the direction of the receiver, i.e., perpendicular to the baseline. The blades turned by 90° to face down, then faced the transmitter, and finally faced up. The signal power seemed to always be of higher intensity when facing the receiver or transmitter but degraded when facing up or down. This signature might be of help in recognizing the drone maneuvering motion in the course of actual drone flight operation.

Anechoic Chamber
In an attempt to minimize other external factors, a controlled environment was used to further confirm the system capability. This experiment was conducted in an anechoic chamber and focused on the feasibility of detecting a default, unwrapped plastic material blade, maintaining the same geometry. Two frequencies (5 and 7 GHz) were used in the micro-Doppler analysis for the experiment in the anechoic chamber. Based on the simulation result, only two higher frequencies were selected because this experiment aimed to validate the theoretical model. Figure 12 illustrates the experimental setup in an anechoic chamber, showing the Tx-to-Target and Target-to-Rx geometry. Following similar scenarios, the blades were placed at the center of the baseline, with 'Facing-Up' and 'Facing-Rx' scenarios, respectively. At first, the detection was conducted with blade wrapped with aluminum foil and later repeated with the default blade without foil.

Doppler Separation between Rotors
An essential component of this work is the detection of a drone through its blade generated micro-Doppler. It is, therefore, of great importance to have an outstanding Doppler resolution that can separate signatures based on the number of rotors used. To achieve this, we considered using 1rotor, 2-rotors, and 4-rotors, respectively, with 5 GHz operating frequency. This may indeed help in

Doppler Separation between Rotors
An essential component of this work is the detection of a drone through its blade generated micro-Doppler. It is, therefore, of great importance to have an outstanding Doppler resolution that can separate signatures based on the number of rotors used. To achieve this, we considered using 1-rotor, 2-rotors, and 4-rotors, respectively, with 5 GHz operating frequency. This may indeed help in identifying a drone from a flapping bird's wing during classification. Similarly, a 15 s integration time with three-states Off-On-Off transitions was observed. The recording started while rotors were in 'Off' position for 5 s, then switched 'On' for another 5 s. The 'Transition' time between 'Off' and 'On' was represented by the projected rise in the signature due to the starting torque from 'Off' to 'On' position. Based on the number of the rotors used during 'On' state, Figure 13 shows a 1-bar for 1-rotor, 2-bars for 2-rotors, and 3-bars for 4-rotors, respectively, manifesting that the Doppler resolution over a reasonable distance was achieved. As always the case, the middle bar of Figure 13c represents rotor R 2 and R 4 as their symmetrical positions make them mapped onto one another.

Plastic Blade Detection
Based on the facts that the default drone blades are either a plastic (usually transparent) or carbon composite material and considering their poor reflective characteristics with conventional radars, we aimed at implementing the FSR geometry coupled with performance enhancement of the chamber, to verify the detection possibility of the default plastic blade by using 5 and 7 GHz frequencies for both Facing-Rx and Facing-Up scenario. By using similar operating frequency, 1-rotor, 2-rotors, and 4-rotors were detected, as shown in Figure 14.  Despite the inverse relation of signal amplitude to the transmitting frequency, the Doppler value improved in the ratio of 10 Hz, as seen in Figure 13. The detection of default plastic blade could be considered successful, despite the loss in resolution and power intensity degradation when compared to blade wrapped with foil.

Actual Drone Detection
An off the shelf commercial drone was used to ascertain its detectability using the FSR geometry. This setup used a parabolic dish antenna as a receiver to increase gain. For a smooth flight condition of the drone, the transmitter was placed on the 11th floor of a tower at engineering faculty Universiti Putra Malaysia (UPM), with 40 m height (H). The transmitter was made to face an open area behind the building during the conduct of the experiment. The transmitter comprised a signal generator and a dual-polarized high-frequency horn antenna with 15 dB gain. The signal generator transmitted 11 GHz frequency with a maximum 15 dBm signal power.
The receiver was set to point towards the direction of the transmitter main lobe. For maximum power received, an effort was made to maintain the line of sight with the transmitter while observing the signal strength by using a spectral analyzer. Figure 15 illustrates the experimental geometry and the aerial view of the experimental site. The receiver consisted of a 65 cm parabolic dish antenna, with a feed horn mounted to point the focal point of the dish, to receive the focused reflected energy by the dish. The receiving antenna had a minimum gain of 36 dBi, 2.6 • beam width at 3 dB points. For our purpose, the low noise block (LNB) circuit inside the feed horn was developed to suit our purpose. At first, we used a high-resolution ADC device to confirm the signal reception before the detection, as shown in Figure 15b. Based on the signal power received of (−47 dBm), we required about 20 dB gain to meet the sensitivity level of the amplitude detector for the down-conversion process and the Doppler signature extractions. To achieve this, a fabricated LNA operating within Ku-band frequencies was, therefore, used during the experiment. We tried as much as possible to imitate the Satellite by using low power transmit and also the actual satellite dish receiver with a narrow beamwidth. detection, as shown in Figure 15b. Based on the signal power received of (−47 dBm), we required about 20 dB gain to meet the sensitivity level of the amplitude detector for the down-conversion process and the Doppler signature extractions. To achieve this, a fabricated LNA operating within Ku-band frequencies was, therefore, used during the experiment. We tried as much as possible to imitate the Satellite by using low power transmit and also the actual satellite dish receiver with a narrow beamwidth.     Figure 15a, the detection was conducted, as described in the subsequent sections.

Target Detection
Before the experimental validation, the main Doppler due to the drone linear motion was simulated by using the geometry in Figure 15a. This helped in having prior knowledge of the expected signature of the target. A target linear motion was simulated with a radial velocity of 16 m/s similar to the actual drone speed. The resultant signature is depicted in Figure 16. Before the experimental validation, the main Doppler due to the drone linear motion was simulated by using the geometry in Figure 15a. This helped in having prior knowledge of the expected signature of the target. A target linear motion was simulated with a radial velocity of 16 m/s similar to the actual drone speed. The resultant signature is depicted in Figure 16.  Figure 16 shows the simulated Doppler signal in FSR due to target linear motion based on the geometry shown in Figure 15a. Both the time domain and STFT representations depicted the phase change at the point the target approached and left the baseline; hence, an expected signature due to FSR geometry was established.
During the experimental work, we had a collection of clutter which was identified as "low clutter" environment. The clutter was further reduced by using a narrow beamwidth of the receiver antenna. Besides, the receiver was Facing-Up (to the sky) with nothing within the coverage area. Figure 15b shows the transmitter, target, and receiver positions and their formation of the FSR geometry. The drone was made to fly up and cross the baseline formed by the Tx-Rx. Immediately, the drone crossed the baseline, it blocked the signal and caused an additional modulation, which would then be propagated to the receiver. The signal, arriving the receiver, was received by the developed LNB of the receiving system. The received signal followed a conventional FSR receiver  Figure 16 shows the simulated Doppler signal in FSR due to target linear motion based on the geometry shown in Figure 15a. Both the time domain and STFT representations depicted the phase change at the point the target approached and left the baseline; hence, an expected signature due to FSR geometry was established.
During the experimental work, we had a collection of clutter which was identified as "low clutter" environment. The clutter was further reduced by using a narrow beamwidth of the receiver antenna. Besides, the receiver was Facing-Up (to the sky) with nothing within the coverage area. Figure 15b shows the transmitter, target, and receiver positions and their formation of the FSR geometry. The drone was made to fly up and cross the baseline formed by the Tx-Rx. Immediately, the drone crossed the baseline, it blocked the signal and caused an additional modulation, which would then be propagated to the receiver. The signal, arriving the receiver, was received by the developed LNB of the receiving system. The received signal followed a conventional FSR receiver processing technique, discussed earlier in Section 3.1, and, therefore, extracted the Doppler signature, as shown in Figure 17. Figure 16 shows the simulated Doppler signal in FSR due to target linear motion based on the geometry shown in Figure 15a. Both the time domain and STFT representations depicted the phase change at the point the target approached and left the baseline; hence, an expected signature due to FSR geometry was established.
During the experimental work, we had a collection of clutter which was identified as "low clutter" environment. The clutter was further reduced by using a narrow beamwidth of the receiver antenna. Besides, the receiver was Facing-Up (to the sky) with nothing within the coverage area. Figure 15b shows the transmitter, target, and receiver positions and their formation of the FSR geometry. The drone was made to fly up and cross the baseline formed by the Tx-Rx. Immediately, the drone crossed the baseline, it blocked the signal and caused an additional modulation, which would then be propagated to the receiver. The signal, arriving the receiver, was received by the developed LNB of the receiving system. The received signal followed a conventional FSR receiver processing technique, discussed earlier in Section 3.1, and, therefore, extracted the Doppler signature, as shown in Figure 17.   Figure 17b. The drone was made to hover and cross the baseline vertically with target-receiver ranges (R rx ) for up to 40 m. The Doppler signature of the drone started as soon as the drone approached the baseline; when the bistatic angle was within the FS mode until the Doppler was approximately close to zero, while the drone was crossing the baseline. All the three-states approaching, crossing, and leaving of the drone signature can be seen in Figure 17, and the phase changes due to target crossing. The difference in the Doppler value between the simulation and the experiment might have resulted from factors, such as variation in the geometrical alignment like the aspect angle, degradation of the real signal due to mechanical noise and surrounding clutter, or affected by volume scattering and background noise. Although detection of a target crossing the baseline was established, yet, a conclusion was not made as to whether the detected target was a drone or not. To further ascertain that, an STFT was used to extract the micro-Doppler features due to blade rotation, as this could serve as an avenue to identifying the drone.

Micro Doppler Extraction
The FSR geometry in Figure 15a was also used to simulate a micro-Doppler signature of a flying drone; this gave an insightful behavior of the combined Doppler, i.e., Doppler due to the main body and micro-Doppler due to rotating blades. As earlier discussed, the micro motions resulted in an additional modulation surrounding the main Doppler. An STFT representation of the simulated signature for one second described the expected performance of the Doppler signature when the drone crossed the baseline. This is illustrated in Figure 18.
The FSR geometry in Figure 15a was also used to simulate a micro-Doppler signature of a flying drone; this gave an insightful behavior of the combined Doppler, i.e., Doppler due to the main body and micro-Doppler due to rotating blades. As earlier discussed, the micro motions resulted in an additional modulation surrounding the main Doppler. An STFT representation of the simulated signature for one second described the expected performance of the Doppler signature when the drone crossed the baseline. This is illustrated in Figure 18. As illustrated in Figure 18, the simulated signature provided a clear view of the Doppler due to the linear motion of the main target body of about 20 Hz and the micro-Doppler surrounding the Vshape signature spread for 1 s. Having this in mind, the experimental result can be compared to also show a combined signature of Doppler due to the drone's body and the micro-Doppler from the rotating blades. The area covered by the signature was also within a period of 1 s as the case of the simulated signature; this is illustrated in Figure 19. As illustrated in Figure 18, the simulated signature provided a clear view of the Doppler due to the linear motion of the main target body of about 20 Hz and the micro-Doppler surrounding the V-shape signature spread for 1 s. Having this in mind, the experimental result can be compared to also show a combined signature of Doppler due to the drone's body and the micro-Doppler from the rotating blades. The area covered by the signature was also within a period of 1 s as the case of the simulated signature; this is illustrated in Figure 19.  Figure 19 shows the experimental signature with micro-Doppler features in the form of spikes surrounding the main Doppler. In both the simulated and the experimental signatures, the Doppler approximately equaled to zero at the point a target crossed the baseline and tended to improve as soon as the drone moved away the baseline forming the V-shape signature of an FSR geometry, as can be seen in both cases. The simulated results complemented the experimental signature, thus confirming the detection of the commercial drone through its micro-Doppler features. Apart from target identification, the micro-Doppler feature could further be used to establish the direction of flight of the drone. In contrast, the Doppler value only reflected the speed lower than the maximum.

Drone Direction of Flight
Here, the effect of micro-Doppler features was used for further identification of other drone movements. A drone performed different movement during flight, such as yaw, pitch, roll, maneuvering, among others, by merely altering the rotating blades mode. For instance, the drone hovered by increasing the net thrust of all the four rotors and performed the opposite when vertically descending to land. During the experiment, the observation time of 30 s was considered. The drone was made to hover after 5 s to cross the baseline height (h), stay up for a while like 20 s, and then descend to cross the baseline for the second time. The intention here was to observe the effect of both  Figure 19 shows the experimental signature with micro-Doppler features in the form of spikes surrounding the main Doppler. In both the simulated and the experimental signatures, the Doppler approximately equaled to zero at the point a target crossed the baseline and tended to improve as soon as the drone moved away the baseline forming the V-shape signature of an FSR geometry, as can be seen in both cases. The simulated results complemented the experimental signature, thus confirming the detection of the commercial drone through its micro-Doppler features. Apart from target identification, the micro-Doppler feature could further be used to establish the direction of flight of the drone. In contrast, the Doppler value only reflected the speed lower than the maximum.

Drone Direction of Flight
Here, the effect of micro-Doppler features was used for further identification of other drone movements. A drone performed different movement during flight, such as yaw, pitch, roll, maneuvering, among others, by merely altering the rotating blades mode. For instance, the drone hovered by increasing the net thrust of all the four rotors and performed the opposite when vertically descending to land. During the experiment, the observation time of 30 s was considered. The drone was made to hover after 5 s to cross the baseline height (h), stay up for a while like 20 s, and then descend to cross the baseline for the second time. The intention here was to observe the effect of both the rotating blades and the drone body, i.e., to say while hovering up; the blades were the first to approach the baseline and vice-versa for landing. Based on this scenario, a 6 s segment of the acquired signal represented a window containing the signature information, the corresponding signatures are illustrated in Figure 20

Drone Direction of Flight
Here, the effect of micro-Doppler features was used for further identification of other drone movements. A drone performed different movement during flight, such as yaw, pitch, roll, maneuvering, among others, by merely altering the rotating blades mode. For instance, the drone hovered by increasing the net thrust of all the four rotors and performed the opposite when vertically descending to land. During the experiment, the observation time of 30 s was considered. The drone was made to hover after 5 s to cross the baseline height (h), stay up for a while like 20 s, and then descend to cross the baseline for the second time. The intention here was to observe the effect of both the rotating blades and the drone body, i.e., to say while hovering up; the blades were the first to approach the baseline and vice-versa for landing. Based on this scenario, a 6 s segment of the acquired signal represented a window containing the signature information, the corresponding signatures are illustrated in Figure 20.  Figure 20a shows the signature of the drone moving vertically up (hovering up) with a higher power intensity and Doppler value when the rotating blades were the first to approach the baseline, and a low Doppler value was observed as it moved away from the baseline. During the landing (Vertically down), the main drone body first approached and crossed the baseline; hence, low power intensity and Doppler value were also recorded, as shown in Figure 20b. These features could be used to identify the direction of flight, and other maneuvered motions by the drone.

Conclusions
This paper highlighted the aspiring benefits of implementing FSR geometry in detecting a target characterized as LSS target like the quadcopter drone. By using a rotating blade, the paper successfully extracted the blade's signature for the number of rotors used, the blade orientation (aspect angle), and the drone maneuvered motion. An effort was also made for proper selection of operating wavelength to make detection easier by considering different operating frequencies. The resulting signature confirmed the assumption of a better signature with wavelength lower compared to the electrical size of the target. Before these, the paper presented a 4-rotor, 2-blade drone model in FSR geometry, which could be used for the micro-Doppler analysis of an LSS. The detection of the default plastic blade without any RCS improvement could be considered as an overwhelming achievement as compared to blades with improved RCS. The detection of an actual Phantom drone for up to an approximate practical distance of 40 m away from the receiver and the micro-Doppler feature extraction due to the blade s micro motions might be a basis to ascertain our claim for drone detection and its micro signature analysis. Also, the benefit of detecting micro-Doppler is the capability to recognize drone from other small targets like birds, etc. Other additional effects of the micro-Doppler features identified, may form the basis of identifying other drone's motions. This study can, therefore, be used to consider an FSR geometry a good option for LSS target detection and classification. The successful inclusion of the parabolic dish antenna as the receiver in the FSR geometry served as an avenue towards the implementation of Digital Video Broadcasting (DVB) satellite-based passive FSR system for drone detection; this has marked our current ongoing project.