Large Scale Simulation of Raw SAR Data for the Sentinel-1 TOPS Mode
Larsen, Yngvar1; Rommen, Björn2; Snoeij, Paul2; Navas-Traver, Ignacio2

The TOPS mode will be the standard mode over land for the upcoming Sentinel-1 mission. The community doesn't have prior experience with this kind of data, so early access to example data is very important. Currently, however, only a few real TOPS mode data sets exist, and none of them are publicly available. To mitigate this, we have developed a versatile raw data simulator with support for all Sentinel-1 modes.
The algorithm is based on an inverse processing approach, where a randomized phase is applied to real SAR backscatter data from a similar geometry, and the resulting complex data are passed through the steps of a SAR processor, but in the opposite order. By starting from a grid with closer pixel spacing than the data product resolution, azimuth aliasing effects may be included at the expense of processing time.
Over land, backscatter statistics from a database based on ASAR GM data, developed by Vienna University of Technology, are used as input. Average wind data from WINDSAT are used together with the CMOD backscatter model over oceans.
The swath dependent nominal elevation antenna pattern of Sentinel-1 is applied to the data. Due to the frequency domain processing algorithm, only the average azimuth antenna pattern is applied in the TOPS mode case. Nominal orbital data provided by the ESA CFI software are used in the simulation, together with the actual Sentinel-1 radar parameters and acquisition timelines. Nominal attitude for Sentinel-1 is assumed. The simulated data are finally compressed with the variable bitrate FDBAQ algorithm and packaged in the Sentinel-1 native SAR Space Packet format.
The correctness of the simulation approach is cross-validated using the Sentinel-1 C++ Prototype Processor, independently developed by Aresys based on a slightly different processing algorithm.
A dataset consisting of an example baseline scenario for a full 12-day repeat cycle (175 orbits), will be simulated. The dataset consists of about 1400 individual data takes of varying length, mode and polarization(s), and the expected raw data volume is 10-15 TB.