A Novel Deep Learning-Assisted Microwave Photonic Direction Finding System Based on Long-Baseline Array

We propose and demonstrate a deep learning-assisted photonic approach for measuring the angle-of-arrival (AOA) with high-precision, which is suitable for long-baseline direction finding (DF). A non-uniform linear array with long-baseline is constructed to increase the precision of AOA estimation and to deal with the problem of ambiguity. The system realizes AOA-to-Voltage mapping by using dual-drive Mach Zehnder modulator (DDMZM) as phase detector and envelope detection in electrical domain. Finally, a deep neural network with long-short term memory (LSTM-DNN) is used for post-processing to establish a mapping relationship between the envelope voltage and real AOA, which not only simplifies the measurement process without phase calibration and transformation between phase difference and AOA, but also compensates the defects of the optoelectronic system and effectively improves the AOA estimation performance. Results obtained using the proposed structure demonstrate less than 0.3405(?) errors over a-80(?) to 80(?) AOA measurement range, and the mean absolute error (MAE) and root mean square errors (RMSE) are 0.1438(?) and 0.3923(?) respectively.

Automated summary

We propose and demonstrate a deep learning-assisted photonic approach for high-precision angle-of-arrival (AOA) measurement suitable for long-baseline direction finding. The system utilizes a non-uniform linear array and a dual-drive Mach Zehnder modulator for AOA-to-Voltage mapping. A deep neural network with long-short term memory is then employed for post-processing to improve AOA estimation performance.