Sentinel-1 New (Thermal) Denoising Approach

Size: px

Start display at page:

Download "Sentinel-1 New (Thermal) Denoising Approach"

Lesley Sutton
5 years ago
Views:

1 Sentinel-1 New (Thermal) Denoising Approach CEOS WGCV 2017, JPL, Pasadena N. Miranda 1, A. Recchia 2, N. Franceschi 2, R. Piantanida 2 1 ESA/ESRIN 2 Aresys S-1 Mission Performance Center: G. Hajduch, R. Husson, P. Vincent, CLS (FR) P. Meadows, A. Pilgrim, BAE Systems (UK) D. Giudici, R. Piantanida, Aresys (IT) D. Small, A. Schubert, UZH (CH) A. Mouche, Ifremer (FR) H. Johnsen, Norut (N)

2 S-1 A/B Noise Equivalent Sigma Zero S-1A IW EW S-1B rg az

3 Impact on data S-1A/B daily mosaics S-1 data and especially the cross-pol is very quickly impacted by the thermal noise This is seen by users as major limitation towards the usage of the cross-pol on many application fields Courtesy of DTU, R. Saldo

4 S-1 Denoising Model The generic signal at the end of SAR processing can be modelled as follows: s r (τ, η) = [s(τ, η) + n(τ,η)] G tot (τ, η) 1,2 The thermal Denoising function on the intensity is given by 1 :! ", $. & ) ) '(' =, -(./0. & 12 & 034 ". & 567 ". & '(4/ $. & 48(9. : -(./0 ) Noise pulse power Unexpected variation of the noise pulse power ICAL Antenna Pattern Spreading loss TOPS descalloping Processor gain Noise calibration 2 Not provided to the users How to best calibrate the denoise vectors ESA 1 Sentinel-1 UNCLASSIFIED Level - For Official 1 Detailed Use Algorithm Definition Detailed-Algorithm-Definition 2 A. Freeman, "SAR calibration: an overview, IEEE TGRS, vol. 30, no. 6, pp , Nov doi: /

Noise Pulse Power variation (σ n2 ) S-1A/B provides noise pulses only at datatake start/end that are

at any time S-1A/B noise power follows a bimodal distribution During the S-1B CP, correlation was

variation to cope with the Earth brightness Tº? 1 Freeman, A. and J. C.

5 Noise Pulse Power variation (σ n2 ) S-1A/B provides noise pulses only at datatake start/end that are separated by several minutes The S-1 processor linearly interpolates between them to get noise power at any time S-1A/B noise power follows a bimodal distribution During the S-1B CP, correlation was found with the Earth brightness temperature 1 Problem: how to capture the variation of noise power variation to cope with the Earth brightness Tº? 1 Freeman, A. and J. C. Curlander, "Radiometric correction and calibration of SAR images, " Photogrammetric Engineering and Remote Sensing, Vol. 55, , Sep

6 Noise pulses for free Postamble For S-1A/B in TOPS, the first rank packets at burst start are not recording any echo (pulses still travelling) Those are noise-like packets that can be used to track the noise power variation over the data-take For S-1C/D real noise pulses will be part of the imaging cycle 11min Noise pulses Preamble Repeated imaging cycle Postamble Preamble IW timeline Noise power variation over the data-take using the Signal free pulses

7 Noise pulses for free but contaminated by RFIs Identification of C band Radio Frequency Interferences from Sentinel-1 data, Andrea Monti Guarnieri et al. to be published

Calibrating the noise vectors (knoise) Denoising vectors are not calibrated and requires to be adjusted in order to reflect the actual NESZ NRCS The noise calibration factor knoise is required to

8 Calibrating the noise vectors (knoise) Denoising vectors are not calibrated and requires to be adjusted in order to reflect the actual NESZ NRCS The noise calibration factor knoise is required to account: - Difference of signal and noise spectral characteristics - Noise power doesn t have a physical unit NESZ estimated over the Doldrums have a peak variation of the order of 2-3dB indicating that (even over the Doldrums) might be contaminated by signal. We need an approach to identify how far the NRCS profiles represents the NESZ such that knoise can be reliably estimated knoise Denoise profiles

Identifying signal free bursts IW3 raw data Reusing previous work

noise estimator OoB power is estimated on each RAW burst after

Out-of-band Power Characterised on-board filter The OoB power can

actual noise pulse power (pre/post amble) * Noise Estimation and

9 Identifying signal free bursts IW3 raw data Reusing previous work from ASAR*, the idea was to assess the Out-of- Band (OoB) power as noise estimator OoB power is estimated on each RAW burst after removing the on-board filter shaping (whitening) In-band Power Out-of-band Power Characterised on-board filter The OoB power can be compared with the: Noise-like power from the first rank echoes actual noise pulse power (pre/post amble) * Noise Estimation and Removal from ASA WSM Products TN, POL ARE TN QWG 04.3 Version Date 10 Nov,2004

10 Example: S-1B EW3 co-pol For co-pol, the OoB power is fully contaminated by signal àin general with S-1, it is very unlikely to find a signal-free burst (over the Doldrums) OoB can t be used as noise power estimate while interleaved noise power remains reliable Each point is a burst 8dB

11 Example: S-1B EW3 cross-pol For cross-pol, good correlation between: Noise pulse power Interleaved noise power Out-of-Band Power Limited contamination of the OoB power by the signal power 0.8dB

12 Identifying signal free bursts Over the Doldrums most of the data is contaminated by signal. Selection needs to be done to estimate accurately NESZ OoB power can t be used as an estimate of the noise power due to the signal contamination However OoB power can be used to identify (especially in cx-pol) the few but reliable signal-free bursts (those closer to the 1:1 correlation line) Out-of-Band power [db] Cross-pol Co-pol Each circle is a burst Echo (in-band) power [db]

Denoise calibration approach Knoise have been

13 Denoise calibration approach Knoise have been calibrated using the few but reliable signal-free bursts Over the Doldrums the error of the denoising vectors is after calibration is <0.5dB for EW and <0.3dB for IW Not clear why IW performs better than EW. Assessment on-going

14 Range intensity profile (natural values) Azimuth intensity profile (natural values) 10dB Thermal denoising with the new vectors Same radiometric scaling

15 Conclusions S-1A/B NESZ performance is better than the -22dB specification The specification is however to high to allow providing noise free data Thermal denoising is necessary to allow effective usage of the cross-polarisation data New approach has been put in place to improve the accuracy of the denoising: Capturing the noise power evolution using the noise-like packets within the imaging cycle to compensate for Earth Brightness Temperature Identifying the signal free bursts to accurately estimate the NESZ and calibrating the denoising vectors Introducing the azimuth denoising in addition to the range already present Those changes will be made effective in the S-1 processor by the Q Further assessment are however required (RFI impact, IW vs. EW performance)

16 THANK YOU

Sentinel-1A Radiometric Calibration

Sentinel-1A Radiometric Calibration Peter Meadows 1, Alan Pilgrim 1, Riccardo Piantanida 2, Davide Riva 2, Nuno Miranda 3 (1) BAE Systems Applied Intelligence, West Hanningfield Road, Great Baddow, Chelmsford,