The Use of Forward Scatter Analysis and AI in Drone Detection.

Prior warning, this one could get quite “techy”.

Many people ask us how our “DRODEC” drone detection systems work, specifically how they capture drones operating “dark” without an outbound Rf signal.

The answer is in a process known as “Forward Scatter Analysis” (FSA). This is then combined with AI in the form of Convolutional Neural Networks to identify specific drone types, manufacturers and payloads. What follows is a fairly technical description of how FSA works in our DRODEC and DeFli Devices.

DRODEC and DeFli devices utilize the advantages of forward scattering radar (FSR) topology and characteristics to detect drones, the empirical mode decomposition (EMD) is applied to the received signal to extract the unique feature vector of the micro-doppler frequency from the drones rotating blades.

Our devices use a signal transmitted from a variety of satellites (VHF, UHF, L-BAND, GNSS) to detect flying drones crossing the forward-scatter baseline between the satellite and the DRODEC/DeFli Device.

Figure 1

The Disadvantages of Current Detection Methods

Detection systems utilizing radio frequency-based (RF), acoustic, video imaging, audiovisual, software-based smartphone, and other non-technical methods like shooting and netting, have a a number of issues that affect performance. Considering the drone-generated audio frequencies that are usually within 40 kHz, drone detection may fail due to a higher noise ratio in urban cities. The drone’s nature of following a predefined GPS route provides no RF link to trace and, as such, cannot be detected. The constant update failure of the drone’s signatures needed by a referenced database in RF-based method and the noisy nature of unlicensed WIFI RF bands due to many users are other challenging issues. In the case of a camera-based method, detection may be hidden due to dynamic target background which, in turn, suffers few pixels representation; these make it difficult to differentiate a drone from flying birds. The thermal technique is considered inefficient due to drone’s plastic frames and minimal heat exhaust.

The Disadvantages of Traditional Radar in Drone Detection

• A drone has a low radar cross-section (RCS) due to its small size, shape, and construction material;

• A drone flying at low speed and erratic flight path is difficult for conventional short-range radar to detect and measure;

• A drone flying at low altitude difficult for conventional short/long-range radar, usually 122 m (400ft);

• A drone has similar behavior to a bird, making it difficult for classification and recognition.

Overcoming With FSR

The key features to detect the drone in the DeFli FSA rely on the received signal’s Doppler frequency and its micro-Doppler scattered by the drones’ blade. The proposed system will enjoy the advantages of:

• High RCS in the main lobe of forward-scatter region. Unlike normal radar, target RCS in FSR is practically independent of the radar absorbing material (RAM) coating, which means it can detect drone with stealth capability;

• Long coherent intervals due to a low drone speed providing a high resolution (∆fFS = 1/∆TFS). This makes it capable of using inverse shadow synthetic aperture for automatic classification/recognition;

• The integration of FSA into passive mode will make the proposed system covert and possibly low cost as there is no need to develop a dedicated transmitter;

• The leakage/direct signal, which is ‘unwanted’ in conventional passive bistatic radar, is good for FSA for target detection due to direct signal perturbation;

• Utilizing multiple frequencies means 24 h signal availability and a large coverage footprint. Thus, the system will have good coverage.

• Does not suffer from environmental effects, such as dark, noisy, and blurred or misty environments.

Micro-Doppler is a vital feature that resulted from an additional frequency modulation generated from the rotation of the drone’s blade. Detecting micro-Doppler is quite challenging due to its non-constant nature, mainly that the drone is engaged in multiple modes while flying. Because of the non-stationary and non-linear nature of the drone motion, the micro-Doppler became a time-varying component, thus traditional time frequency-processing techniques are non-optimal to extract micro-Doppler. For this reason, we use an improved signal processing technique based on empirical mode decomposition (EMD), to extract the unique feature vector from the received signal.

An FSA is a special bistatic mode when a bistatic angle is near to 180°. In FSA geometry, the target crossing between the transmitter (Tx) and receiver (Rx) blocks the electromagnetic (EM) wave travelling from the Tx-to-Rx; this forms a target’s shadow (silhouette) irrespective of the target’s material. The FSA minimizes the effect of radar cross-section (RCS) of the target while detecting. The high target’s RCS irrespective of the target’s shape and radar absorbing material (RAM) makes it popular over the traditional monostatic radar; instead, FSA considers the target physical dimension and the wavelength during detection. The FSR system has an improved feature of being counter to stealth technology, low RCS targets detection, high power yield, among others. These identified benefits may fit its usage in detecting a target characterized as a low altitude, slow speed, and small RCS target like a quadcopter commercial drone.

Signal Reception

The rotating parts of a drone may have micro-motions referenced to the main silhouette-generating center of the target’s body; this results in additional frequency modulations of the reflected wave and, hence, micro-Doppler . The signature is now determined by the additional motion of the target that is different from the main body’s motion. The micro-motion may be caused by rotation, vibrations, or both , resulted from multiple combinations of point scatterers.

The FSA 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.

Empirical Mode Decomposition

To extract some feature vectors from the original signature of the detected target we use the Empirical Mode Decomposition (EMD) technique. We use the extracted vectors as a stronghold for identifying the detected drone in FS-mode, based on the peculiar signature of a Doppler pattern with V-shape characteristics (towards zero Doppler on the center of baseline) and micro-Doppler surrounding the main Doppler. This is also used in the identification of this target among other targets in the same bandwidth or surveillance volume, like flying birds. The empirical mode decomposition (EMD) involved a sifting process capability to decompose any given signal into an intrinsic component and derived non-linear functions called Intrinsic Mode Functions (IMF) This technique is a successive algorithm that enables the primary function, of which can be non-linearly derived from the original signal in an adaptive basis called the IMF. These IMFs are the corresponding frequency components available in the original data, efficiently helping to separate between the Doppler due to radial velocity and the micro-Doppler due to the rotating blade.

AI and Deep Learning

We use Convolutional Neural Networks (CNN’s) for additional object detection. A CNN is used to classify and regress the input Range-Doppler Radar (RDR) maps patch and the Euclidean distance between the patch center and the target. A non-maximum suppression (NMS) technique was proposed to reduce and control the false alarm, this mechanism can detect the target more precisely and attain a much better false alarm rate than conventional constant false alarm rates (CFARs) for radar detection of moving targets.

The use of IP protected deep learning-based software within the DRODEC / DeFli devices / systems tracks, detects, and classifies objects from raw data in real-time using the convolutional neural network technique. Due to their comparatively high accuracy and speed, deep convolutional neural networks have been proven to be a trustworthy method for picture object detection and classification. In addition, a CNN method enables UAVs to transform object data from the immediate surroundings into abstract data that machines can understand without human intervention. Machines can make real-time decisions based on the facts at hand.The ability of a UAV to fly autonomously and intelligently can be significantly enhanced by integrating CNN into the onboard guiding systems.

Contact us

Whether you have a request, a query, or want to work with us, use the form below to get in touch with our team.