Technical Specification

Impact of Radio Frequency Interference on C-band

Since the start of the Sentinel-1 mission, users have reported local image degradations related to radio frequency interferences (RFI). Known sources of RFI include: the Canadian RadarSAT system, ground radars and weather radars operating at the same wavelength. Compared to previous spaceborne SAR missions, S-1 suffers from interference not only within specific conflict zones, but worldwide. Hence, there is a strong need for a mapping and classification system of RFI in the C-band.

To perform thermal noise estimation, Sentinel-1 acquires blocks of signal-free echoes in the initial and final section of each data take. Recently, ESA increased the number of noise measurement available, also including echoes sensed at the beginning of each data burst. Since March 2018, then, signal-free measurements in the C band are available with a rate of about 1Hz. It was shown [RD1] that these echoes can be used to extract relevant statistics for performing RFI detection.

The RFI global mapping tool

An RFI mapping tool was developed in the context of an ESA-funded project to extract the relevant information from the Sentinel-1 L0 Noise Products. This tool implements the techniques described in [RD1] and its flow diagram is shown in Figure 1.

A calibration step is needed to estimate the shape of the on-board decimation filter and the position of the spurious frequencies found in the range frequency spectrum. The output of this calibration routine is then used in the estimation phase, when, for each L0 Noise product, the power spectral density of the rank echoes is whitened and multi-looked.

Flow diagram of the RFI detection tool

The RFI detection step is performed using two statistics, the one-tailed Fisher’s Z test:

\[Z = {H_f – E[H_f] \over VAR[H_f]}.\]

where \(H_h\) is the spectrum of the data; and the Kullback–Leibler (KL) divergence:

\[\int p(H_h)log ( {p(H_f) \over p_{fit}(H_f)} ) dH_f \]

with \(p_{fit}\) being the Normal distribution. The two statistics must be used together on the data because they have complementary objectives: Fisher’s Z finds isolated RFI peaks that deviate from the average power level; KL divergence measures the distance between the power distribution of the noise and the Normal distribution, so it can capture RFI at low power spread on a larger bandwidth.

The RFI database

The RFI sources identified by the detection algorithm and annotated in the RFI DB are characterized according to multiple criteria. The RFI characterization is fundamental for the definition of the best mitigation strategies. The following criteria are considered:

  • Power: RFI power is the parameter directly measured from the L0N products. High power RFIs are a big nuisance because they can “blind” the instrument preventing any mitigation strategy. The digital number recorded by the S-1 instrument was converted to a brightness temperature value using JAXA’s AMSR-2 sensor as an external source of calibration.
  • Frequency: the frequency of the detected RFI sources can only fall in the bandwidth sensed by the instrument corresponding to 5405±150 MHz (the bandwidth varies with the mode of observation, so observations of RFI in the full 100 MHz band cannot be ensured). In the L1 SAR images, the RFI sources falling outside the acquired swath bandwidth are heavily attenuated by the on board decimation filter and, if still visible, aliased in the swath bandwidth.
  • Bandwidth: the sources can be divided into narrow and large bandwidth. The large bandwidth sources usually present reduced peak energy in the spectral domain but can only be mitigated in the time domain. On the other hand narrow band sources usually have higher peak energy and can be mitigated directly in the frequency domain.

The first three parameters are directly annotated in the RFI DB in terms of central frequency, power (brightness temperature) and bandwidth. The characteristics of the detected RFIs can be easily visualized using the interactive maps generated by the RFI tool.

Table 1 display the database structure and the data format.

VariableDB NameUnit
Acquisition TimetimeUTC time string
Sensor IDsensorString [SENTINEL1A / SENTINEL1B]
Swath IDswath_idString [IW1 / 2 / 3/ EW1 / 2 / 3 / 4 / 5]
Polarizationpolarizationstring [VV / VH / HH / HV]
Orbit Directionorbit_directionstring [ASCENDING / DESCENDING]
Center Frequencycenter_frequencyHz
Fisher’s Zfisher_zDimensionless
KL divergenceklDimensionless
Latitude latitudeDeg
Longitude longitudeDeg
Power powerDimensionless digital number
Brightness Temperaturebrightness_tempK
Table 1 Database structure and data format

Sentinel-1 Noise calibration

The calibration problem

The estimated RFI power needs to be calibrated before performing the de-noising operation. This need is well motivated by Figure 2, where the histograms of the S-1A noise power are shown for different antenna beams and polarizations. It was observed that the two platforms (A and B), the two RX polarizations (V and H) and the eight different antenna beams (IW1/2/3, EW1/2/3/4/5) yield distributions differing both in mean and in variance.

Traditionally, the S-1 mission performance center has performed noise calibration empirically: the beam- dependent and polarization-dependent \(k_{noise}\) factors are obtained aligning the noise vectors annotated in the S-1 products to the data profiles over regions with no backscatter (cross-pol data over the Doldrums). The main limitation to this approach is that the assumption that the backscatter of cross-pol Doldrums data is at the same level as the Noise Equivalent Sigma Nought (NESZ) is not always valid.

Aligning the noise levels of different beams and sensors is beneficial to produce significant global brightness temperature maps and give physical meaning to the recorded RFI measures. As different antenna beams exhibit different noise power level, the first activity consists in compensating for the known sources of miscalibration, including noise bandwidth and attenuation of the SAR electronic sub-system. Secondly, to translate the S-1 noise levels to brightness temperature measures, a source of external calibration is chosen: the AMSR-2 microwave radiometer.

Noise power after relative

Brightness temperature after AMSR2 calibration

External calibration using AMSR2 data

The power values \( \hat{P} \) measured by the sensor can be modeled as:

\[\hat{P} = P_n + P_e\]

where \(P_n\) is the effective noise power and \(P_e\) is the power radiated from the Earth surface and captured by the S-1 antenna. The term \(P_e\) can then be converted into the so called brightness or radiance temperature, representing the temperature a black body in thermal equilibrium with its surroundings would have to be to duplicate the observed intensity of a grey body object at a certain frequency. The Earth brightness temperature is the quantity usually returned by spaceborne passive radiometers.

To eliminate the residual bias among the antenna beams and give physical significance to noise measures, S-1 noise data was cross-calibrated using AMSR-2 radiometer aboard JAXA’s GCOM-W1. There are some differences between the two sensors: both of them are in C-band but AMSR-2 operates at 6.93GHz, while S-1 at 5.4GHz. This 1.53GHz difference should not result in significant brightness temperature differences, as the Earth emissivity is quite stable around those frequencies. Both orbits are sun-synchronous, but the local ascending time is 06:00 AM for the S-1 and 1:30 PM for AMSR-2.

The cross-calibration was performed using AMSR2’s L3 products from July 1 2019 and S-1 L0N products of the same day. An AMSR2 L3 consists of four daily brightness temperature grids: one for each polarization (H and V) and one for ascending and one for descending passes. All the grids have a resolution of 0.1 degrees in latitude and longitude. Each S-1 noise measurement from July 1 2019 was collocated with the closest point in the AMSR-2 grid. Then, a linear regression was run to find the linear relation between them. Since different S-1 polarizations and beams are expected to produce different brightness temperatures, a line was fit for each configuration.

The RFI Map

Starting from June 2019, two maps are produced for each S-1 orbit cycle (12days) of each sensor: an RFI probability map and an interactive RFI map showing the main events.

The RFI probability map is obtained dividing the number of RFI events recorded in the area by the number of noise lines analysed in that area. The RFI probability maps have to be considered in the context of the S-1 acquisition plan (the Earth surface is not observed homogeneously).

The interactive RFI map shows the main occurrences detected during the corresponding cycle and can display additional information by clicking on the circles. The radius/colour of the circles is proportional to the brightness temperature of the RFI.


[RD1]Monti-Guarnieri, A.; Giudici, D.; Recchia, A. Identification of C-Band Radio Frequency Interferences from Sentinel-1 Data. Remote Sens. 2017, 9, 1183.
[RD2]S1-RS-MDA-52-7441, Sentinel-1 Product Specification, issue 3.3, October 2016
[RD3]ST-ESA-S1QC-ICD-002, S1QC: Sentinel-1 Quality Control: QCSS Internal ICD
[RD4]ST-ESA-S1QC-DPM-001, S1QC: Sentinel-1 Quality Control: QCSS Detailed Processing Model.