SENTINEL-1 FDBAQ PERFORMANCE VALIDATION USING TERRASAR-X DATA Elke Malz Ilmenau University of Technology Electronic Measurement Lab 98693 Ilmenau elke.malz@tu-ilmenau.de Rolf Scheiber Josef Mittermayer German Aerospace Center DLR Microwaves and Radar Institute 82230 Oberpfaffenhofen rolf.scheiber@dlr.de Paul Snoeij Evert Attema European Space Agency ESA- ESTEC 2201 Noordwijk paul.snoeij@esa.int ABSTRACT Two Block Adaptive Quantization (BAQ) algorithms considered for implementation on-board Sentinel-1, the Entropy Constrained BAQ (ECBAQ) and the Flexible Dynamic BAQ (FDBAQ) are investigated with real data acquired by TerraSAR-X. The two algorithms are compared with respect to the resulting signal-to-quantization-noise ratio (SQNR) and the compression rate. The results confirm the improved performance of FDBAQ to be expected for Sentinel-1 compared to the more conventional ECBAQ. Index Terms—Block adaptive quantizer, quantization, data compression, synthetic aperture radar (SAR). 1. INTRODUCTION SAR satellite data provide a wide range of applications. Technically the recent SAR sensors provide large bandwidth, high sensitivity, multiple operating modes and increased operation time in orbit. Future digital beam forming promises much wider swaths leading to a further increase of the payload data rate. This forces a better on-board raw data compression since the data downlink remains often the bottleneck of a satellite SAR mission. The conventional data compression technique applied to SAR systems is often the block adaptive quantization BAQ [1], but future SAR missions require methods with better compression performance. Beside vector quantization methods [2] and Trellis coding [3] there are also effective BAQ algorithms, like the Entropy Constrained BAQ (ECBAQ), [4,5] or the newly introduced Flexible Dynamic BAQ (FDBAQ), [6,7,8] selected for implementation on-board Sentinel-1 [6]. In this paper the anticipated FDBAQ performance in terms of SQNR and compression rate based on simulation is confirmed using real satellite data and compared to ECBAQ. Essential for this investigation was the availability of high dynamic range (8- bit) data from TerraSAR-X. 2. ALGORITHM DESCRIPTION 2.1. Entropy Constrained Block Adaptive Quantization The Entropy Constrained BAQ is a widely discussed BAQ algorithm in the field of Synthetic Aperture Radar. There are a lot of variants of this algorithm, which are e.g. described in [4, 5]. Within this work we have used the algorithm as it is described in [4], which consists of a uniform BAQ followed by an Entropy coder combined with a step size control. An Entropy coder is a lossless coder; here we use a Huffman coder. The BAQ coding is performed in the time-domain. The step size control corresponds to a scaling of the input data of the individual block to a reference standard deviation ref σ . The combination of a uniform BAQ with a Huffman coder was shown to be the optimum quantizer [4, 5]. The approximate average distortion achievable by this combination is comparable to the Shannon optimum performance, as given by the distortion rate bound [5]. The block diagram of the ECBAQ is given in Fig. 1. Inputs are the block length BL, necessary for the extraction and recombination of the individual blocks, as well as the bitrate R BAQ , which is fixed for the whole acquisition. Fig. 1. Block diagram of the ECBAQ. 2.2. Flexible Dynamic Block Adaptive Quantization The FDBAQ was first introduced in [7]. It is a BAQ which can have uniform or non-uniform spacing between thresholds followed by an Entropy coder, see [7, 8]. Since the described ECBAQ already defines an optimum quantizer, for the FDBAQ the same BAQ kernel as for the ECBAQ has been implemented. In contrast to ECBAQ, the FDBAQ adapts the bitrate of every block with respect to the range location of the data block and thus with respect to the NESZ variation along range, and with respect to the mean signal backscatter, by fulfilling the so- called NESZ boundary condition. Thus, it uses a minimum number of bits and establishes a uniform NESZ along range to quantize the data. In Fig. 2, the block diagram of the FDBAQ is shown. The range dependencies are determined by generation of the so-called range dependent scaling function () r η . The range dependent scaling function is sensor dependent and given in Section 3 for the implementation with TerraSAR-X real test data. Given optimized quantization thresholds (defined for uniform (or non-uniform) quantization according to the maximum/minimum tolerable SQNR) and a backscatter model are used to calculate the minimum bitrate, which fulfills the NESZ boundary condition. According to the 5 discrete bitrates uniform BAQ sd Entropy Coder raw data block length BL raw data block bitrate RBAQ compressed raw data block ... ... compressed raw data ECBAQ