Multi-Radar Interference Mitigation in Photonics-Based Radar With Sliding Window LSTM Recurrent Neural Network

Hum Nath Parajuli, Aigerim Ashimbayeva, Ukash Nakarmi, Ikechi Augustine Ukaegbu, Bijay Gaudel, Shilong Pan, Carlo Molardi, Bikash Nakarmi

2024

Closed access Metadata (CSL-JSON)
DOI 10.1109/jlt.2024.3402329
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

In this paper, we propose and experimentally demonstrate a technique of mitigating multi-radar interference in a photonic radar using a sliding window long short-term memory (SW-LSTM)-based recurrent neural network. For the proof-of-concept experimental demonstration, a photonics-based victim radar (VR) with an 8 GHz linear frequency modulated (LFM) signal is generated using optical injection in a semiconductor laser. This VR is used to detect two objects separated by 10 cm in the presence of multiple interference sources. To perform the analysis of mitigating interference in the multi-radar environment, we generated random interference LFM signals (IR-LFM) with diverse bandwidths (4 GHz to 10 GHz) and chirp rates (±1 GHz/μs to ±20 GHz/μs), both experimentally and synthetically. The generated IR-LFM signals consist of both coherent and non-coherent interferences and are added to the echo signal and trained using SW-LSTM. During the training stage, learning rate, stacked layers, nodes, batch size, and window size are optimized. In the test stage, we use signals with both coherent and non-coherent interferences, which are distinct from the interferences in the training stage. The proposed model is evaluated for one, four, and eight interference test scenarios, in which it successively achieves a meanR2value of > 93% for all test scenarios, taken over ten trials in each test scenario. Furthermore, a mean error vector magnitude(EVM)ofSINR) of > 11 dB are obtained. These results indicate a strong similarity between the range profile calculated in the interference-free and the multi-interference environments with our proposed scheme of mitigating interference using SW-LSTM.