FR2752062A1 - METHOD FOR ENCODING RAW SAR DATA WITH DATA REDUCTION - Google Patents

Abstract

The invention relates to a method for encoding raw SAR data with data reduction, the encoding being performed on the basis of correlated data produced by means of a reversible transform (T).

Description

1 2752062

  The invention relates to a method for coding raw SAR data.

with data reduction.

  For economical storage and transmission of SAR data, their relatively large data rate should be reduced by means of a

source coding.

  Studies carried out on the statistical properties of SAR data have shown that lossless entropy coding of raw data brings almost no data reduction [2] (JC Curlander and RN McDonough: Synthetic Aperture Radar. John Wiley & Sounds, INC. 1991). On the other hand, coding methods affected by losses such as vector quantization have made it possible,

  in accordance with [3] (J.M. Moureaux et al .: Vector quantization of raw SAR data.

  In: Proceedings of the International Conference on Acoustics, Speech & Signal Processing, vol. V, pp. 189-192, IEEE., 1994) to obtain a compression factor of the order of 2 to 4. It remains to be seen how far the losses

  of information that has occurred leads to perceptible disturbances of the image.

  Since some applications do not admit losses

  Of information, lossless coding methods are of particular interest.

  The object of the invention is therefore to indicate a method for coding raw SAR data which allows coding without losses or affected by losses, with

  controllable losses for comparatively increased coding efficiency.

2 2752062

  This object is achieved by the fact that the coding is carried out on the basis of correlated data produced from the raw SAR data by means of a

reversible transformation (T).

  The method according to the invention uses, for coding, raw SAR data suitably transformed. To obtain a reduction as much as possible without loss of the raw SAR data rate, a reversible transformation is used (hereinafter referred to as T). The purpose of the transformation is to transform the almost uncorrelated raw SAR data into correlated data which can then be coded using image coding methods well known per se. The more strongly the data are correlated, the more efficient the well known image coding methods work. The correlated transformed data correspond approximately to the reflexivity distribution of the

land surface.

  After the encoding of the transformed data, these are stored and respectively transmitted and decoded. Using the inverse transformation T'-, the raw SAR data is reconstructed without losses. The recovered raw SAR data can be transmitted to a traditional SAR processor for image calculation. Used in a SAR system on board a satellite, the method presented can provide savings in channel capacity for transmission or, on board a

  aircraft, memory capacity for storage.

  The invention is described in more detail using examples.

  FIG. 1 represents the entire system for processing raw SAR data according to the invention; Figure 2 shows the SAR coordinate system used; FIG. 3 represents different stages of the transformation T; FIG. 4 represents a concrete embodiment of the transformation T. FIG. 1 shows the entire system for processing raw SAR data. The actual raw SAR data received by the SAR antenna is first demodulated in an HF receiver (not shown here). As a result of the demodulation, we obtain the formation of raw raw SAR data which serve as input signal of the T transformation. The application of the T transformation leads to the formation of correlated data which are then coded in a coder at help

3 2752062

  image coding methods well known per se. The data thus coded can be transmitted (for example from a satellite to the ground station) or stored in a memory (for example in a SAR system on board an aircraft). For the recovery of SAR data, the data is decoded in a decoder and reconstructed without loss with application of the reverse transformation T- '. Fig. 2 represents the typical SAR geometry in the case of a SAR system on board an aircraft. The antenna fixed on the airplane moves at the flight speed VA in the direction x (azimuth). The distance from objects on the surface

  of the Earth is characterized by z (oblique distance).

  The shortest distance between the SAR detector and the surface of the Earth is designated by zo. The complex raw SAR data s (x, z, t) can be described as follows: s (x, O, t) = I + jQ (1) with I = 91e {s (x, O, t)} and Q = Zm {s (x, O, t)}, (2) where I (in-phase) and Q (quadrature) are the so-called quadrature components of the signal s (x, z, t), ot is the propagation time and z = 0, the place of exploration in the

SAR detector.

  The transformation T is carried out in two stages TR and TA, as indicated in FIG. 3. The technical realization of the transformation stages TR and

TA is illustrated in fig. 4.

  The first step of the transformation T, the distance transformation TR essentially consists of a modified pulse compression. To ensure reversibility, a modified pulse spectrum PM (CO) is used for the transformation of distance TR. The modified pulse spectrum PM (co) differs from the real pulse spectrum P (wo) in that outside the limit frequency og of the real pulse spectrum, the value of the modified pulse spectrum is kept constant, in accordance with IP (0) l for I o I <rg IPM (O) l = {C = const. otherwise (3) The limit frequency o) g is of the order of half the bandwidth

  of the pulse used in the SAR system.

4 2752062

  The phase of the modified pulse spectrum is equal to the phase of the real pulse spectrum J PM (0) = <P (o) (4) IPM (co) I = C has the effect of avoiding possible disturbing influences caused by the fact that in the transformation TR, we multiply with very low numerical values, and that in TR-, we divide by the same low numerical values.

  The distance transformation TR provides the signal d (x, 0, t).

1 0

  TR {S (X, 0, t)} = f S (x, 0, o) P * M (oe) eiJtdoe = d (x, 0, t). (5) 2ic J

  * then means complex conjugate.

  Fig. 4 (a) shows the technical realization of equation (5). TR is performed according to a one-dimensional Fourier transformation in the direction t, as a multiplication with P * M (c0) in the frequency range. The inverse Fourier transformation provides the distance transformation data d (x, O, t)

wanted.

  The TA azimuthal transformation of raw SAR data is derived from a technique introduced in [1] (C. Cafforio, C. Prati and F. Rocca: SAR data focusing using seismic migration techniques. IEEE Transactions on Aerospace and electronic Systems, vol. 27, pp. 194-206, March 1991) as a new focusing method for SAR data. The TA transformation consists of two filters for all frequencies. Filtering for all frequencies is done in the frequency range. For this, we still need two one-dimensional Fourier transformations as well as an inverse Fourier transformation

two-dimensional, see fig. 4 (b).

  After the Fourier transformation of the range transformation data d (x, 0, t) in the direction x, d (x, 0, t) o --- x D (kX, 0, t) (6) the first filter for all frequencies Hl provides with the transfer function C k2x H1l (k, t) = exp (j -, (7) 4 kb the signal D '(kx, 0, t) = D (kx, 0, t) * Hl (kx, t). (8) After the application of a second Fourier transformation in the direction t to the result D '(kx, 0, t) following D' (kx, O, t) o - -, D (kx, O, (o), (9) the second filter for all frequencies H2 provides, with the transfer function Ct0 H2 (kx, kz) = exp [-j (- / k2. + (Kzo + kz) 2 - k20z, - kzZo)] r (10) the signal D '(kx, k., - to) = D' (k., O, co) H2 (k., kx). (11) Regarding equation 10, this is the exact result for any bigle angles of the SAR antenna. In particular for the simulations of the method according to the invention on the computer, it is necessary to make the approximations following in the case of small bigles angles: zok2x zok2x zok2xKz H2 (kx, k) exp - () exp - (-)] (1) 2 (kzo + kz) 2kzo0 2k22o The two-dimensional inverse Fourier transformation D '(kx ,, kz, -to) -obkkZ d (x, z, -to) (13) provides the desired transformed and correlated data. They correspond substantially to the reflexivity distribution of the earth's surface at the time of the

3 0 reflection -to.

  For the recovery of the raw data, a reverse transformation TA- is first carried out. Since the value of Hl (kx, kz) and H2 (kx, t) in equations 8 and 10 is constant and equal to unity for all frequency components, the inverse transformation TA-I consists of a multiplication of the transformed data with the same exponential terms, only with the inverted sign of the exponents. The inverse transformation TR can be written as inversion of equation 5 in the form of 1 c0 D (x, 0, (0) TR-1 {d (x, 0, t)} = - J ejot dco = s ( x, 0, t) (14) 27c JP * M () -00o The transformation T described ensures a numerically stable behavior of the transformations TR, TA as well as their inverses. Measurements of the signal-to-noise ratio bring out the fact that the transformation allows an exact reconstruction of the raw SAR data in case no quantification of the transformed data is made. Following the transformation T, coding of the

  transformed data, with data reduction.

  With the realization of the reversible T transformation, we made available a tool which allows efficient coding with data reduction. An important advantage of this method is that the coding can now be performed on correlated transformed data. Measurements of the entropy of the real and imaginary components of d (x, z, t = -to) demonstrate that it is possible, without taking advantage of the correlation, to perform data reduction without loss with the

compression factor of approximately 2.

  Instead of entropy coding, it is now also possible to implement other lossless image coding methods which exploit, in addition, the correlation with data reduction. As examples, one can cite here 1. Transformation coding such as the discrete cosine transformation (DCT) followed by the entropy coding of the transformation coefficients

  2. Partial band coding followed by entropy coding.

  3. By admitting image coding methods affected by losses, it is possible, instead of or in combination with the methods cited above, to resort to

  also to vector quantization.

7 2752062

  The system for coding raw SAR data according to fig. 4 can be realized with integrated programmable signal processes. To carry out the transformation T, the algorithms described must be implemented in the software. For consecutive encoding and decoding, it is possible to do

  use of commercially available integrated circuits for image coding.

  These should however be adapted to the statistical properties of SAR data.

raw processed.

8 2752062

  Bibliography [1] C. Cafforio, C. Prati and F. Rocca: SAR data focusing using seismic migration techniques. IEEE Transactions on Aerospace and Electronic Systems, vol. 27, pp. 194-206, March 1991. [2] J.C. Curlander and R.N. McDonough: Synthetic Aperture Radar. John Wiley &

Sounds, INC. 1991.

  [3] J.M. Moureaux et al .: Vector quantization of raw SAR data. In Proceedings of the International Conference on Acoustics, Speech & Signal processing, vol.

V, pp. 189-192. IEEE, 1994.

Claims (7)

  1. - Method for coding raw SAR data, since the coding is carried out on the basis of correlated data produced from   raw SAR data by means of a reversible transformation (T).   2. - Method according to claim 1, characterized in that the transformation (T) comprises a distance transformation (TR) followed by a azimuthal transformation (TA).   3. - Method according to claim 2, characterized in that the distance transformation (TR) comprises a modified pulse compression according to () I for I? (cW) for {ca 1 - <cog IPM (C = {C = const otherwise o PM (c) is the real pulse spectrum and og, an appropriately chosen limit frequency. 4. - Method according to claim 2 or 3, characterized in that the azimuthal transformation (TA) comprises two filtering for all the frequencies with the transfer functions C k2x Hl (kx, t) = exp (j - t), 4   2 o 4 kzo cto H2 (kxkz = exp [-j (-4V k22 + (ko + k2) 2 - kozo - kazo)] 5. - Method according to claim 4, characterized in that the second filtering for all frequencies alternatively presents the following transfer function: zok2, zok2x zok2xkz H2 (kx, kz); exp [-j ()] exp -j (- -)] 2 (k2o + kz) 2k ,, 2k2ZO   6. - Method according to one of the preceding claims, characterized in   what the coding performed on the basis of the correlated data is an entropy coding.   7. - Method according to one of claims I to 6, characterized in that   the coding performed on the basis of the correlated data is a transform coding followed by an entropy coding of the transformation coefficients.   8. - Method according to one of claims 1 to 6, characterized in that   the coding performed on the basis of the correlated data is a partial coding followed by a entropy coding.   9. - Method according to one of claims 1 to 6, characterized in that   1 0 the coding carried out on the basis of the correlated data is affected by losses and includes vector quantization.