4.8 Ascending and Descending Scenes Coregistration
The research explores SAR data acquired from both ascending and descending orbits. In some instances, these scenes, or products derived from them, are merged together into a single image. As briefly explained in Section 2.1.2, foreshortening, layover and shadow effects in the radar slant direction can have a drastic impact on an image especially in the presence of steep topography. Due to the different points of view of the ascending and descending scenes, these distortion effects are produced in different directions and consequently the combination of such data sets must be carried out with particular attention. This matter has not been significantly investigated in the literature. The fusion of multi-angle ERS-1 and 2 tandem data has been applied in previous research (Sansosti et al., 1999; Coltelli et al., 1998 and Carrasco et al., 1997) to improve the accuracy of digital elevation models and to detect ground displacements in three dimensions as opposed to a single-orbit-direction tandem pair which can instead only detect displacements in the SAR slant direction. More specifically, Werner et al. (2002) discuss issues related to SAR geocoding and multi-sensor image registration. They use the satellite orbit and local terrain information to reduce coregistration errors. They also select recognisable image features as ground control points as an additional aid. Their analysis highlights the problem of selecting certain persistent scatterers which might not have the same position in the ascending and descending images due to their height (e.g. high buildings). The utilisation of SAR images with different acquisition geometries has also been applied by Dell’Acqua et al. (2003). They manually coregister three simulated SAR images with 3 m ground resolution and generated each with a different viewing geometry. Dell’Acqua et al. (2003) try to extract an accurate map of the road system of the city of Pavia in northern Italy and do not use the images for a direct multi-band classification because the same buildings have a difference appearance in the two images.
The relativistic phase and time offsets from this paper are not only of high importance for DEM generation with a formation flying SAR cross-track interferometer. Formations with multiple satellites have also been suggested for a wide range of further remotesensing applications, ranging from along-track interferometry for moving object and ocean current measurements over sparse aperture ambiguity suppression and super resolution for enhanced high-resolution wide-swath SAR imaging up to single-pass SAR tomography for vertical struc- ture measurements –. Due consideration of relativistic effects from varying along-track baselines is again of essential importance for these advanced bistatic and multistatic SAR systems to avoid mutual range and phase offsets between the received SAR signals. The phase accuracy requirements for the combination of the different receiver signals are typically in the order of 1° or a few degrees. For comparison, an along-track baseline of 100 m causes in an X-band system a relativistic phase shift in the order of several tens of degrees. Future multistatic SAR satellite missions should therefore take into account relativistic effects in the design of the radar synchronization system and/or the SAR processor to avoid a possible performance loss.
A number of algorithms for focusing the raw SAR data have been developed since its debut in 1950s. Most of them were originally developed forremotesensing app- lications. Therefore some approximations and assumptions due to the restrictions of the hardware, e.g. antenna beamwidth, squint angle, the computation ability of signal processor, have been made when focusing the raw SAR data. The most accurate SAR algorithm is the time-domain correlation algorithm , which can be used to pro- cess the SAR data acquired with arbitrary beam-width and space sampling trajectory. Neither restriction on the antenna’s beamwidth nor motion compensation need to be applied. However, it requires a very high computational effort, which makes it im- practical for any true real-time SAR application given the performance of a realistic hardware environment [43, 44]. By restricting the space sampling interval to uniform, frequency-domain algorithms, which perform the focusing in frequency domain with the power of FFT, can substantially reduce the computational burden.
Symmetry assumptions about the distribution of ele- mentary scatterers within the resolution cell simplify the scattering problem and reduce the number of independent parameters of [ ] T (or [ ] C ) allowing qualitative and quanti- tative conclusions about the scattering behavior , , . Besides reciprocity, three special cases of symmetry are important in radarremotesensing applications: Reflection, rotation and azimuthal symmetry. Reflection symmetric media are characterized by a symmetry plane that contains the line-of-sight so that for any scatterer located at one side of the plane a mirrored scatterer at the other side of the plane exists. In this case the correlation between the co- and cross- polarized elements becomes zero. The resulting [ ] T matrix contains only five independent parameters in form of three real diagonal elements and one single non-zero complex off-diagonal element (i.e., the correlation between the co- polarized elements). The majority of natural distributed scatterers is reflection symmetric. In the case of rotation symmetry, the spatial distributions of elementary scatterers do not change when rotated about the line-of-sight (LOS) axis. Accordingly, the scattering behavior of such media is invariant under the line-of-sight rotations and the resulting coherency matrix contains only three independent parame- ters in form of two independent real diagonal elements and one non-zero imaginary off-diagonal element. This is typi- cal for gyrotropic random media, as given for example by a random distribution of helices. When both, reflection and rotation symmetry applies, the medium is said to be azi- muthally symmetric: All planes including the line-of-sight are reflection symmetry planes. Consequently, all three off- diagonal elements of the coherency matrix become zero, and only two diagonal elements are independent, the num- ber of independent parameters reduces to 2. This is the case for volumes consisting of random distributions of ellipsoids.
A big challenge for future spaceborne remotesensing missions is now turning to the estimation and long-term monitoring of dynamic processes in the Earth’s environmental system, such as deformation events, forest and biomass change, and ocean surface cur- rents. The German Aerospace Center (DLR) is investigating an innovative single-pass interferometric and fully polarimetric L-band radar mission, named Tandem-L, which ex- ploits innovative high-resolution wide swath SAR modes, together with the use of large bandwidths, high pulse repetition frequencies, and multiple acquisition channels, result- ing in an achievable swath width of about 350 km on ground. Such an increase in term of coverage has as a main drawback the generation of a huge amount of onboard data, which is of around 8 Terabytes per day. One of the proposed solutions to reduce the resulting onboard data reduction suggests to perform a complex onboard processing (i.e. an onboard interpolation, low-pass filtering and decimation) and allows a data reduction up to 50% . On the other hand, the onboard computational memory required for the data reduction processing is at the limit of the hardware components, leading to high energy consumption. Moreover, the practical realization of the technique is very complex, including many specific coefficients which must be correctly selected during acquisition. The research of alternative solutions is therefore of great interest in order to have differ- ent options to choose for the mission development, which motivates the present master thesis. In this work, a data reduction strategy based on Linear Predictive Coding (LPC) is investigated in the context of Tandem-L. The method has been designed to reduce the complexity as much as possible while achieving a certain data reduction. A mathemati- cal formulation for the novel technique is an interesting goal for understanding in which situation the present method can be more or less efficient. The resulting performance has been verified through Monte Carlo simulations in order to evaluate the solution under different aspects (i.e. final performance versus resulting system complexity). Moreover, other complications introduced by the Tandem-L system such as the presence of missing samples (so-called gaps) during the acquisition, have been investigated and successfully solved through novel coding strategies.
This paper has summarized new techniques and concepts towards a vision of a constellation of SAR satellites for earth remotesensing. Bi- and multi-static SAR configurations in combination with digital beamforming make optimum use of the total available signal information and energy, allowing an efficient radar design for applications with user requirements for frequent monitoring, wide-swath imaging, interferometry and high resolution. Digital beamforming with multiple aperture signals allows for the efficient suppression of ambiguities, which enables new SAR systems with wide coverage and high image resolution. This avoids conflicts from operating SAR systems in mutually exclusive imaging modes such as scanSAR, stripmap and spotlight and enables regular observations of large areas. Further potential advantages of bi- and multi-static radar systems in combination with digital beamforming are enhanced MTI, efficient interference suppression, resolution enhancement and SAR tomography.
The application of the classical displaced phase centre (DPC) technique in a high resolution wide swath (HRWS) SAR system puts a stringent requirement on the pulse repetition frequency (PRF): the PRF has to be chosen such that the SAR platform moves just one half of the to- tal antenna length between subsequent radar pulses. This requirement is due to the fact that the phase centres of the additional samples from the multi-aperture antenna must exactly fit in between the satellite’s travelled distance be- tween two transmit pulses to yield a uniformly sampled syntheticaperture. However, such a rigid selection of the PRF may be in conflict with the timing diagram and it will furthermore exclude the opportunity to increase the PRF for improved azimuth ambiguity suppression. To avoid such restrictions, a new reconstruction algorithm has been developed in  which allows for the recovery of the unambiguous Doppler spectrum even in case of a non-uniform DPC sampling. The derivation of this recon- struction algorithm is based on modelling the raw data acquisition in a multi-aperture SAR by a linear multi channel systems model where each channel describes the SAR data acquisition of one antenna element. The recon- struction assumes a limitation of the Doppler bandwidth B Dop ≤ N*PRF, where N is the number of receiver chan-
The spaceborne active remotesensing have the unique capability of observing sea surface through cloud, which has been playing an important role of monitor- ing response of sea surface under extreme weather situations. Scatterometers on board the European RemoteSensing (ERS), the Quick Scatterometer (QuikSCAT), and the Meteorological Operational (MetOp) satellites are particularly suitable for measurements of sea surface wind field, as both wind direction and wind speed can be derived without needing external information. Another active remotesensing instrument, spaceborne SyntheticApertureRadar (SAR), e.g., the ERS-1/2 SAR, ENVISAT/ASAR, RADARSAT-1/2, TS-X/TD-X and Cosmo-Skymed, can not only provide sea surface backscatter intensity like scatterometer, but also image the sea sur- face in two-dimension with large spatial coverage and high spatial resolution, which provides abundant oceanic and atmospheric information of TCs, such as hurricane- generated long swell waves in small scales [ 3 , 4 ], hurricane/typhoon eye morphology [ 5 ] and roll vortices occurred in marine boundary layer [ 6 ] in meso-scale.
SyntheticApertureRadar (SAR) is an active remotesensing technology. Radar satellites emit and receive their own signals and, hence, do not need to rely on sunlight for their acquisitions. Another benefit compared to optical satellites is SARs ability to look through clouds, which means data acquisition is very reliable.
Typically, EO satellites are placed in Low Earth Orbit (LEO), to take advantage of the shorter slant range. The key limitation of LEO forremotesensing is the difficulty of providing frequent timely images of an area. A single satellite typically images a swath 100-200 km wide steerable within a field of view 500 km across. LEO orbit periods are approximately 100 minutes, and in this time, the Earth has turned almost 3000 km at the equator: it thus takes ~3000/500=6 days to obtain complete access. Envisat actually uses a 35-day repeat orbit, which means a repeat period of 5 weeks for interferometry, and 1-2 weeks (at mid-latitudes) for imaging. Processes with timescales shorter than these periods are difficult to measure usefully. Satellites in geosynchronous orbit, on the other hand, have a permanent view of 1/3 of Earth’s surface so that, as soon as one image is complete, the next can be started. Allowing for the time required to acquire the image this means that a geosynchronous satellite could provide in principle several images each day of any location in the field of view. An inconvenience caused by the geosynchronous orbit is its fixed longitude. This implies that only a constellation of at least three satellites can guarantee a nearly global coverage. Even so, polar zones cannot be imaged with nominal resolution. However, global coverage is not always an essential requisite for a space system. The problem that many satellite service providers have to tackle is that the request for satellite products is concentrated in the most industrialised regions of the globe therefore a system that is able to provide nearly continuous imaging of a certain region may prove to be advantageous for certain customers.
The selection of the spatiotemporal excitation coefficients for the individual Tx apertures could even be made adaptive by evaluating the recorded samples from previous signal returns (cf. Fig. 3). By this, a closed loop will be formed between the radar sensor and its environment, which allows for a maximiza- tion of the information that can be derived about the imaged scene for a given RF power budget. In analogy to the informa- tion theoretic modeling of multiple-input multiple-output (MIMO) communication systems, such an optimization could then be regarded as maximizing the mutual information be- tween the recorded radar signals and the scatterer distribution on the ground, thereby making optimum use of the channel capacity provided by the multiple antenna Tx/Rx radar system. For illustration, one may consider the simple case of an auto- matic compensation of angular variations in the received Rx power being caused by, e.g., range differences, inhomogeneous atmospheric RF signal attenuation, and/or spatial variations in the first-order scattering statistics of the imaged scene.
If spaceborne syntheticapertureradar (SAR) sensors were categorized according to their operational flexibility, four generations could be identified. The first sensors were on/off sensors with a single fixed mode of operation (fixed beam, in- cidence angle, bandwidth, etc.) Later, sensors were developed which could be operated in multiple modes, such as StripMap, ScanSAR or SpotLight; however the macro-based command- ing was basically restricted to selecting a specific mode without the flexibility to alter/access individual instrument set- tings. Current SAR sensors –third generation– offer a greater flexibility in commanding nearly each individual parameter of the instrument (see T ERRA SAR-X  for example). This can be understood as providing the basic building blocks to construct any operation mode in combination with any possible instrument setting. However, even current sensors do not offer a satisfactory solution to the “fundamental limitation of SAR sensors” which can be described in short as the incapability to simultaneously provide high resolution and wide coverage , instead they offer the flexibility to choose a compromise between high resolution or wide coverage.
channel by combining the signals from overlapping subgroups of multiple receiving antenna elements prior to A/D conversion and storage. This, again, enables better sidelobe suppression and further allows a dynamic adaptation of the position of the effective phase centers, thereby improving the signal re- construction . In principle, one could also increase the number of azimuth channels while keeping the total antenna length fixed. This raises the number of samples of the syn- thetic aperture and, hence, enables more efficient suppression of residual ambiguities. A drawback of this solution is, of course, the increased data rate to be stored on the satellite and to be transmitted to the ground. The data volume can, however, be reduced onboard the satellite by a multichannel data compression that exploits the mutual redundancies (here, mainly second-order cross correlations) between the signals recorded from neighboring antenna elements. Optimum data compression algorithms and architectures can be derived from information theory by employing a rate distortion analysis , which has to take into account the desired performance of the final image product. This can be regarded as the more general view of the previously introduced preshaping of the receiver beams, and it will ultimately lead to the design of digital radar architectures, which are significantly driven by elements from information theory. 3
Whereas the predictions for a correction with the translation approach have already been validated for a PS-InSAR time series (B¨ ahr et al., 2012), the reference frame effect itself has not been explicitly observed yet. Nevertheless, its existence is evident without explicit verification, and a correction is advisable whenever large-scale deformation phenomena are measured with high accuracy requirements. Based on the comparison in figure 7.8, a dedicated Euler rotation of orbital state vectors can be considered sufficiently accurate in most cases. Also a homogeneous translation derived from one representative ITRF velocity performs well but may involve minor phase artefacts. It should be preferred nevertheless in regions where the approximation quality of plate kinematic models is bad. The general transformation approach may be worth considering whenever InSAR measurements are used to densify an existing (GNSS) velocity field. If the benefit of the complex transformation outweighs the involved effort in this case still needs to be investigated.
The challenge of this approach is to obtain a good 3D position accuracy of the phase center of the antenna. The green dashed curve (b) in Fig. 2 shows a sinusoidal deviation in y-direction from the ideal flight path (red solid (a)) with an amplitude of ±50 mm. This deviation is known. The blue dotted curve (c) shows an unknown normally distributed deviation of the ideal flight path with an amplitude of ±10 mm. The corresponding radar data and the processed SAR images are shown in Fig. 3. As long as the motion of the UAV is pre- cisely known, the back-projection algorithm can compensate
Index Terms— GEO SAR clutter, Parametric land clutter model, SAR clutter, wheat clutter model
Geosynchronous SyntheticApertureRadar (GEO SAR) has attracted increasing interest in the last two decades. The concept is now widely accepted, but there are some concerns on the performance achievable on non-static target scenes. The movement of a target in a SAR system causes target signal to be smeared in the azimuth direction. This smeared signal is a form of clutter. For some GEO SAR mission con- cepts, the azimuth spread of the power scattered from clutter is a potentially important constraint on imaging performance because it smears a noise-like power across the image. Mod- els of GEO SAR imaging therefore need to include clutter spread [1, 2].
SAR sensors commonly use wideband chirp signals to illuminate targets perpendicular to the moving direction of the sensor platform. With the long transmit pulses in range and the wide antenna beams in cross-range direction, the received raw data is initially defocused. Refocusing is done in a SAR processor that uses the exact knowledge of the system transfer function and the platform motion. With the SAR principle that is based on a coherent system, the parameters must be available as complex numbers i.e. in magnitude and phase. The radar antennas are anisotropic as a function of spatial direction and demand for a three dimensional characterization in an azimuth over elevation coordinate system over frequency. The F-SAR processer that is as well developed at the Microwaves and Radar Institute allows 3D antenna pattern correction. Compensating for the antenna is a computationally intensive step in SAR processing that requires precise, three dimensional antenna measurements.