GNSS-R Delay/Doppler Map Compression Method Using a Denoising Convolutional Autoencoder

To reduce the channel transmission and data storage in the spaceborne Global Navigation Satellite System Reflectometry (GNSS-R), a Delay/Doppler Map (DDM) compression method for CYGNSS products is proposed in this paper. It applies a self-supervised denoising convolutional autoencoder (DCAE) to dig out effective features in the DDM, which are used to reconstruct the DDM. This network is an encoder-decoder symmetric structure, which can reduce about 90% of the data volume. CYGNSS L0, L1a, and L1b DDMs are all used to validate the compressing performance in this paper. Results show that the averaged root mean square errors (RMSEs) between reconstructed and original L0, L1a, and L1b DDMs are 2816 counts, 2.8*10
⁻¹⁸
W and 5.7×10
¹⁰
m
²
respectively. The averaged peak signal-to-noise ratios (PSNRs) are respectively 19.48 dB, 13.67 dB, and 6.79 dB. The encoder-decoder structure of DCAE can effectively compress and restore DDMs. We also verify the effectiveness of features from the encoder by a simple fully connected network (FCN) for wind speed retrieval. The RMSE of wind speed retrievals is 1.76 m/s, which proves that compressed features carry roughness information.