Fusion of Deep Representations in Multistatic Radar Networks to Counteract the Presence of Synthetic Jamming

Abstract

Micro-Doppler signatures are extremely valuable in the classification of a wide range of targets. This work investigates the effects of jamming on micro-Doppler classification performance and explores a potential deep topology enabling low bandwidth data fusion between nodes in a multistatic radar network. The topology is based on an array of three independent deep neural networks (DNNs) functioning cooperatively to achieve joint classification. In addition to this, a further DNN is trained to detect the presence of jamming and from this it attempts to remedy the degradation effects in the data fusion process. This is applied to real experimental data gathered with the multistatic radar system NetRAD, of a human operating with seven combinations of holding a rifle-like object and a heavy backpack which is slung on their shoulders. The resilience of the proposed network is tested by applying synthetic jamming signals into specific radar nodes and observing the networks' ability to respond to these undesired effects. The results of this are compared with a traditional voting system topology, serving as a convenient baseline for this work.

Human Micro-Doppler Signature Classification in the Presence of a Selection of Jamming Signals

“…A key advantage the jamming system has over a victim radar is that jamming signals only travel half the distance of the radar signal. To achieve comparable powers at the radar receiver, a jamming signal can have a power proportional to the victim radar's transmitting power divided by the square of the radar to target range [9]. Jamming systems will generally be at a disadvantage in most other aspects.…”

Evaluating Effect of Blanket Jamming on Radar Via Robust Time-Frequency Analysis and Peak to Average Power Ratio

“…In practice, by associating lemma 4, lemma 5 and equation 30together, we have the following theorem: Theorem 2: For PD radar with LFM signal and blanket jamming, assume that N r and N a are its sampling numbers on range and azimuth, respectively. If the SJR of x 0 in (24) is SJR 0 and the PAPR of 1, µ ≈ 0. For conciseness, the proof of the theorem above is omitted.…”

“…This limitation arises from the fact that they seldom put emphasis on jamming mechanism and thus fail to uncover the universal rules on effect of jamming on radar. Although the data-driven methods, e.g., convolutional neural networks [23], [24] and principle competent analysis [25], are utilized to address this issue by exploring the information hidden in the data (namely samples), this hurdle has not been completely tackled due to their black-box property.…”

Deep Fusion for Radar Jamming Signal Classification Based on CNN

“…Recent studies have shown that the deep neural networks (DNNs) have strong representation ability for complex structures and excellent performance in the field of radar feature fusion. For example, Patel et al [17] investigated the effects of jamming on the micro-Doppler classification performance and explored a potential deep topology based on DNNs enabling low-bandwidth data fusion between nodes in a multistatic radar network. Xiang et al [18] designed a novel direction of arrival estimation method based on DNN for very high-frequency radar under strong multipath effect and complex terrain environment, which can learn the received data's characteristics from a different elevation.…”