2.6.1.2.4.2 Descalloping

The purpose of this section is to discuss the problem known as scalloping, and the process used to correct it, known as descalloping.

2.6.1.2.4.2.1 Scalloping

Since a sensor operating in burst mode does not view a scene continuously, achieving uniform image quality over the imaged extent is more difficult for a burst mode imaging system than for a continuous mode one. In particular, burst mode images are subject to cyclic radiometric banding effects as described below.

During azimuth compression, the data in each burst must be adjusted by the proper azimuth gain correction function. This is due to the fact that the return energy from a single scatterer is modulated in azimuth according to that portion of the antenna gain pattern used in the burst imaging event, an effect known as scalloping Ref. [2.11 ] (see figure2.34 Below). In other words, different point targets are illuminated by different positions of the azimuth gain pattern yielding different integrated powers. The amount of scalloping corresponds to the energy difference from the beginning to the end of the processing bandwidth and is generally measured in decibels (dB).

Figure 2.34 Return Energy From a Single Scatterer (3 bursts shown)

However, an inaccurate estimation of the fractional Doppler centroid frequency 2.6.1.2.2. leads to a misapplication of each of the azimuth descalloping functions to its corresponding burst image return signal. As a result, the output signal level is rendered non-constant across the azimuth burst as shown in figure2.36 a below. In this figure, the Doppler centroid error ( f _Derror ) causes a cyclic rotation in the uncompensated burst energy function along azimuth (as shown in the solid line), resulting in an incorrect application of the descalloping function (shown as the dotted line).This radiometric variation in azimuth across each burst results in a cyclic radiometric pattern known as residual scalloping across the full image in azimuth, as depicted in figure2.36 b.

Figure 2.35 The application of a suitable correction function to an azimuth burst with an accurate Doppler centroid frequency estimate

Figure 2.36 The application of a suitable correction function to an azimuth burst in the presence of a Doppler centroid frequency estimation error

2.6.1.2.4.2.2 Methods for Scalloping Correction

2.6.1.2.4.2.2.1 Azimuth Descalloping Function Application

Inverse Beam Pattern Descalloping Method One method to correct for scalloping across azimuth involves applying weighting functions which are inversely proportional to the azimuth antenna gain function. This technique shall be referred to as the Inverse Beam Pattern Method for descalloping.

To apply the correction effectively, the Doppler centroid frequency must be known accurately.

The weighting functions derived using the Inverse Beam Pattern method emphasise the outer looks to equalise the power, thereby maximising the equivalent number of looks, or speckle reduction, over azimuth.

Constant signal-to-noise ratio (SNR) Descalloping Method Bamler's "Optimum Look Weighting for Burst-Mode and ScanSAR Processing" ( Ref. [2.11 ] ) has developed a set of antenna pattern correction functions for burst mode and ScanSAR processing. The functions are derived to satisfy the following criteria in the multi-look azimuth processing case:

• The image signal energy becomes constant over azimuth,
• The noise energy becomes constant over azimuth, and
• The equivalent number of looks (which describes the amount of speckle reduction obtained by a weighted multi-look process) is maximised over azimuth.

In addition, the constant SNR method generates weighting functions which de-emphasise the outer looks, so that the equivalent number of looks is not maximised.

2.6.1.2.4.2.2.2 Doppler Centroid Frequency Estimation in the context of Burst Mode Data

2.6.1.2.4.2.3 Notes

2.6.1.2.4.2.4 References

2.6.1.2.4.3 ScanSAR Beam Merging

ScanSAR multi-beam processing consists of generating independent beam images (called beam buffers) and subsequently combining them into a single output line, as shown in figure2.37 below:

Figure 2.37 Merging Beam Buffers

Data in the merged line is either copied directly from one beam, or the merged line data is a blend of pixels from the so called blend region of the two beams.

Therefore, the beam merging algorithm consists of three primary steps described ago.

2.6.1.2.4.3.1 Determine a reference point in the blend region

2.6.1.2.4.3.2 Determine the limits of the blend region

The nominal size of the blend region, say N (even number), is specified in a PF-ASAR input parameter file. A number of N/2 points are extracted on either side of the reference point for each (near and far) beam, so the reference point is precisely the centre of the blend region. If necessary, N will be clipped against the limits of the overlap region.

Figure 2.38 The Blend Region and its Reference Point

2.6.1.2.4.3.3 Merge the data in the two beams corresponding to the blend region according to a predefined rule

In describing the method, the following notation is used:

xnear(n) = the array of near beam pixel power values;

xfar (n) = the array of far beam pixel power values and;

p = weight rate

For a given value of p, the formula used to calculate the merged pixels is given by:

xmerged(n) = (1 - (n/N)p) * xnear(n) + ((n/N)p) * xfar(n),

where n =0,1,...,N-1 and p >= 0. The data on the steeper extremities of the beam pattern, where the effects of a bias in the elevation beam pattern correction are more significant, are disregarded while the data on the steeper extremities of the blend region are weighted less by the merging model. This is because the following continuity conditions hold true:

xmerged(0) = xnear(0)

xmerged(N) = xfar (N)

In particular, if xfar(n) = xnear(n) for all n = 0,1,...,N-1, then

xmerged(n) = xfar(n) = xnear(n) for all n = 0,1,...,N-1

The desired merging scenario can be easily selected by simply setting the weight rate parameter, p, as shown in the following table:

2.6.1.2.4.4 Global Monitoring Mode Inverse Filter

The transmitted pulse typically has a rectangular amplitude, and a quadratic phase variation that gives the pulse its linear FM character. For linear FM pulses, increasing the duration of the pulse allows greater frequency variation across the pulse, thus increasing the pulse bandwidth and resulting in a higher resolution after pulse compression. (See the description of range compression in the section on Range-Doppler 2.6.1.2.3.1.2. processing). Linear FM pulses that are used for pulse compression typically have a large time-bandwidth product, at least 100.

For linear FM pulses with a large time-bandwidth product, there is a time-frequency relationship, such that the amplitude spectrum has the same shape as the pulse amplitude in the time domain. For example, the figure below shows the amplitude of the Fourier transform (FFT) of a large time-bandwidth product pulse with a rectangular shape.

Figure 2.39 Pulse spectrum with large time-bandwidth product.

The Fourier transform (FFT) also has a quadratic phase variation that needs to be removed during matched filtering, in order to compress the pulse. In the frequency domain, the matched filter is defined as the conjugate of the Fourier transform of the pulse. For large time-bandwidth product pulses, multiplication of the pulse spectrum by its conjugate removes the quadratic phase variation, and leaves the amplitude approximately unchanged because of the rectangular shape, resulting in the sinc shaped impulse response.

However, in Global Monitoring (GM) Mode where the required resolution is quite low, a relatively short pulse with smaller bandwidth is used. In this case the time-bandwidth product is quite low, on the order of 20, and the time-frequency relationship does not hold as well. For example, the figure below shows the amplitude of the Fourier transform of a pulse used in Global Monitoring Mode. The amplitude is no longer rectangular, and multiplication by the matched filter would worsen the amplitude variation, which affects the shape of the compressed pulse.

full size

Figure 2.40 Pulse spectrum with small time-bandwidth product.

To solve the problem caused by the shape of the amplitude spectrum when the time bandwidth product is low, the variation of the amplitude spectrum is removed by multiplying the pulse spectrum by the inverse of the Fourier transform (IFFT)of the pulse, rather than the conjugate. This removes the amplitude variation of the spectrum during pulse compression, improving the shape of the compressed pulse.

2.6.1.2.5 ASAR Strip Line Product Processing

2.6.1.2.5.1 2.6.1.2.5.1 Introduction

The ASAR Instrument may be operated in one high rate mode continuously (an acquisition segment) for up to ten minutes duration or for up to one orbit duration in a low-rate mode called Global Monitoring (GM) mode. Stripline products contain image data from an entire acquisition segment, up to a maximum of 10 minutes per product for IM, AP, and WS and up to a full orbit (100 minutes) for GM. All medium-resolution and browse products (IMM, IMB, APM, APB, WSM, WSB, GM1, and GMB) are processed as stripline products. The browse products are subsampled versions of the medium-resolution products and are processed at the same time. ( For a further discussion of all of these products refer to the section entitled "Level 1B Image Products" 2.6.2.1.1.3. and "Browse Products" 2.6.2.1.3. , which can both be found in chapter 2 ).

The PF-ASAR will systematically generate output parent products, within the time constraints, that cover an entire acquisition segment by processing a number of separate slices and then joining the slices to form the continuous segment. Most long data segments will be processed using more than one processing computer, in order to meet throughput performance requirements. Each computer will be given a portion of the input data segment to process. The output, referred to as slices, from the different computers are then concatenated to produce one long strip product, referred to as stripline product.

This stripline parent product is stored as a single product in the Archive Facility (ARF) and key information about the product is maintained in the inventory database for access by users. After the segment product is archived, a customer may extract a subset, or child product. If a customer orders a floating scene from within the stripline product the appropriate portion of the archived product, i.e. the measurement data, annotation data, Specific Product Header (SPH) and Main Product Header (MPH), will be extracted to form the requested output product. The extracted subset product may start at any point in the segment and shall not contain any radiometric or geometric discontinuities.

Geometric continuity is achieved by ensuring all PF-ASAR computers establish the same output grid. All PF-ASAR will use the same input parameters (orbit data, earth ellipsoid parameters, centre swath elevation angle, etc.) and the same software to establish the output grid. The projection used for stripline products differs from the normally used for SAR products. SAR products are usually projected such that the first pixel in range is kept at a constant range. For stripline products, however, the ground range projection is designed to allow very long azimuth extent images to be displayed without cutting off any data and with a minimum of black-filled pixels. This is done by keeping the mid-ground range pixels at a constant elevation angle.

Radiometric continuity is achieved by ensuring the Doppler Centroid frequency is continuous along the data segment. The input raw data to the different PF-ASAR computers is overlapped and each PF-ASAR uses the overlap regions (at the beginning and end of the slice) to estimate the Doppler Centroid frequency. The Doppler Centroid frequency that is used within the slice is calculated by interpolation.

The slice size and the overlap depend on the mode (IM, AP, WS or GM). The figure and table below show the minimum and maximum slice sizes and the minimum overlap. The overlap must take into account the amount of data required for Doppler centroid estimation and processing throw away (matched filter throw-away and azimuth skew).

Figure 2.41 Slice Size and Overlap Illustration

• Slice size
• Slice number
• Acquisition start time of the segment
• Overlap size

2.6.1.2.5.1.1 Slice Size and Overlap

2.6.1.2.5.2 Processing Parameters Affected By Strip Line Product Continuity Requirements

Table 2.35 below "Processing Parameters Affecting Stripline Products", identifies processing parameters that are important for stripline processing and groups the parameters depending on their source:

• signal data - parameters derived from the SAR raw data;
• auxiliary data - parameters derived from common auxiliary data provided to all ASAR computers;
• ISP header data - parameters extracted from header of downlink data and may vary throughout data segments.

2.6.1.2.5.2.1 I And Q Channel Statistics

Channel statistics are calculated from raw data to determine I and Q channel bias and orthogonality adjustments. These adjustments affect radiometric accuracy. ERS experience indicates that channel bias values are relatively slowly varying, therefore statistical analysis of data from any portion of the data segment could be applied to the entire segment, or the analysis could be repeated using any portion of the segment and yield similar results.

Each PF-ASAR computer evaluates I and Q channel statistics from its own ISP data. The raw data used for statistics is a subsample of the total raw data taken from samples throughout the input data. The calculated correction factors are applied uniformly for the entire slice.

2.6.1.2.5.2.2 Chirp Replica

The chirp replica is derived by processing the periodic calibration data. Periodic calibration activities performed by the instrument generate calibration data for segments of the SAR antenna in a cyclic manner and downlinks this data interspersed with the imaging data. The pulse replica is derived from the summation of calibration data from all 32 antenna segments received over a period of time.

During stripline processing the chirp replica is updated whenever a new periodic calibration cycle is found in the input data. If a complete set of calibration pulses from 32 antenna segments is not found, then the nominal chirp is used.

2.6.1.2.5.2.3 Antenna Elevation Gain Correction

The antenna elevation gain correction is also derived by processing the periodic calibration data and is also updated whenever a new periodic calibration cycle is found in the input data. If a complete set of calibration pulses from 32 antenna segments is not found, then the nominal antenna elevation gain correction is used.

The antenna elevation pattern is converted to slant range pattern so it can be applied to the range compressed data. This conversion is updated periodically to follow the continuously changing satellite/earth geometry.

2.6.1.2.5.2.4 Doppler Centroid Estimation

If the overlap data is missing, Doppler centroid estimation is still done using the data at the beginning and end of the available input data. In this case, the Doppler centroid may not be continuous across the adjacent slices. This is acceptable because there is no valid image at the boundary of the slices.

The Doppler centroid frequency maximum variation is about 1.7 Hz/sec. Over a slice of 20 seconds, the Doppler centroid frequency changes by a maximum of 34 Hz and the variation can be approximated as a linear function over a long period.

The ambiguity estimate is performed once per slice and is then applied throughout that slice.

( For a further discussion of the Doppler centroid frequency, ambiguity, and estimation, see the section entitled "Doppler Centroid Frequency Estimator" 2.6.1.2.2. ).

2.6.1.2.5.2.5 along-track Pixel Spacing/Start Position

The Line Time Interval in seconds (the grid spacing along-track) is pre-defined for each product type and subswath and is a constant throughout the orbit to ensure repeatability of grid spacing from one orbit to the next. (It should be noted that the precise pixel spacing in metres will vary depending on the instantaneous velocity of the satellite.)

Each PF-ASAR calculates the zero Doppler start time of the stripline product segment by converting the acquisition time to zero Doppler time.

Each PF-ASAR then calculates the zero Doppler start time and stop time of its output slice.

2.6.1.2.5.2.6 Slant Range/Ground Range (SR/GR) Conversion To Output Grid

As the satellite altitude varies with time, the mid swath line moves closer or farther away from the subsatellite track and the valid data follows the Sampling Window Start Time (SWST).

The number of samples per range lines in the output product is set to be large enough to cover the shift of the valid data along the segment so that no valid data is thrown away.

The valid samples are projected to the output grid using an SR/GR conversion table. PF-ASAR calculates the SR/GR conversion table periodically along the azimuth direction and performs linear interpolation between adjacent SR/GR conversion tables so that a smoothly varying SR/GR can be achieved.

2.6.1.2.5.2.7 FM Rate

2.6.1.2.5.2.8 First Output Range Sample

2.6.1.3 Level 1B Accuracies

The following two tables define the product geometric accuracies of the Level1B products.

In the tables, Pixel means pixel spacing, Resolution means spatial resolution, ENL is equvalent number of looks (which affects radiometric resolution). The ENL on some images is dependant on the instrument setting, indicated with one asterisk (*). Browse images are derived from Medium Resolution images.