EXPERIMENTAL STUDY OF RADIO FREQUENCY INTERFERENCE DETECTION ALGORITHMS IN MICROWAVE RADIOMETRY

1
EXPERIMENTAL STUDY OF RADIO FREQUENCY
INTERFERENCE DETECTION ALGORITHMS IN MICROWAVE
RADIOMETRY

• José Miguel Tarongí Bauzá
• Giuseppe Forte
• Adriano Camps Carmona
• RSLab
• Universitat Politècnica de Catalunya

2
Introduction

• Radio Frequency interference (RFI) present in
  radiometric measurements lead to erroneous
  retrieval of physical parameters.
• Several RFI mitigation methods developed
• Time analysis
• Frequency analysis
• Statistical analysis
• Time-Frequency (T-F) analysis
• Short Time Fourier Transform (STFT) 1
• Wavelets 2
• STFT combines information in T-F, useful if
  frequency components vary over time.
• Spectrogram ? image representation of the STFT.
• Image processing tools can detect RFI present in
  a spectrogram.

1. Tarongi, J. M Camps, A. Radio Frequency
Interference Detection Algorithm Based on
Spectrogram Analysis, IGARSS 2010, 2010, 2,
2499-2502. 2 Camps, A. Tarongí, J.M. RFI
Mitigation in Microwave Radiometry Using
Wavelets. Algorithms 2009, 2, 1248-1262. c
3
Introduction
Frequency analysis
Time analysis
Spectrogram analysis
4
Hardware Settings

• RFI detector hardware
• Microwave radiometer based on a spectrum analyzer
  architecture
• Composed by
• L-band horn antenna G -17dB _at_ 1.4 1.427GHz
• Chain of low noise amplifiers 45dB Gain and
  1.7dB Noise figure
• Spectrum analyzer able to perform Spectrograms
• Calibration and temperature control unnecessary
• Only used to detect RFI
• Measurements taken in the
• Remote Sensing lab from the UPC

RFI detector Schematic
5
Algorithm description

• Objective ?gt Image processing tools applied to
  the spectrogram to detect RFI.
• 1st idea use algorithms previously developed 1
• Pixels conforming the spectrogram obtained by the
  spectrum analyzer have a Raileigh distribution
• Frequency response of the RFI detector hardware
  was not sufficiently flat
• New algorithm developed
• 2D wavelet-based filtering to detect most part of
  the RFI
• Frequency and time averaging to eliminate the
  residual RFI

1. Tarongi, J. M Camps, A. Radio Frequency
Interference Detection Algorithm Based on
Spectrogram Analysis, IGARSS 2010, 2010, 2,
2499-2502.
6
Algorithm description

• 1st part, 2D wavelet based filtering
• Convolution with two Wavelet Line Detection (WLD)
  filters
• WLD filters matrixes based on a Mexican hat
  wavelet
• Two different filters
• Frequency WLD (FWLD) detects sinusoidal RFI.
• Time WLD (TWLD) detects impulse RFI.
• Values of these filters
• FWLD TR rows (15 TR 31), each
• one composed by the coefficient values
• of a Mexican hat wavelet of 11 samples
• TWLD TC columns (15 TC 31), each
• one composed by the coefficient values
• of a Mexican hat wavelet of 11 samples
• RFI enhancement with the correlation of FWLD and
  TWLD with the spectrogram

Mexican Hat coefficient values
7
Algorithm description

• 1st part, 2D wavelet based filtering
• Threshold to discriminate RFI in both filtered
  spectrograms
• Function of the standard deviation of the
  RFI-free noise power ( ) which must be
  estimated
• WLD threshold (TWLD or FWLD)
• Threshold selected to have a Pfa lower than
  510-4
• 1st part of the algorithm can be performed
  several times.

K constant to determine the Pfa ci ith
coefficient of the mexican hat wavelet (11
samples) N of rows/cols of the FWDL/TWDL
filtered spectrogram
with
8
Algorithm description

• 2nd part, frequency and time averaging
• After 2D wavelet filtering it still remains
  residual RFI, next pass
• Average of the frequency subbands
• Average of the time sweeps
• Spectrogram matrix is converted in two vectors.
• RFI is eliminated with threshold proportional to
  the standard deviation of both vectors
• Threshold selected to have a Pfa lower than
  510-3

9
Algorithm description
RFI cleaned signal power
Spectrogram
FWLD filter
TWLD filter
2D Convolution
?


2nd pass RFI mitigation result
FWLD threshold
TWLD threshold

Frequency threshold
Time threshold

1st pass RFI mitigation result
Yes
No
Any frequency subband or time sweep with
relatively high power (6 times above sfreq or
stime) value?
Frequency subbands Time sweeps average
10
Results

• Measurements performed at the UPC (D3-213 bldg)
• L-band (1.414 - 1.416 GHz)
• Continuous sinusoidal wave and impulsional RFI
  detected
• Sinusoidal RFI Vertical lines
• ImpulseRFI Horizontal lines

• Spectrogram of a radiometric signal in the
  "protected" 1.400 - 1.427 MHz band with clear RFI
  contaminated pixels.
• Vertical line CW RFI at 1415.4 MHz
• Horizontal line Impulsional RFI at 36 s

11
Results
TWLD filtering and thresholding
FWLD filtering and thresholding
12
Results
threshold
Time averaged spectrogram
threshold
Frequency averaged spectrogram
13
Results
Frequency and time averaging
2D wavelet based filtering
14
Conclusions

• Best RFI algorithm is actually a combination of
• 2D image filtering of the spectrogram using line
  detection filters.
• Time and frequency analysis to the remaining
  radiometric signal
• System equalization may be performed
• Avoid false alarms from the RFI detection
  algorithm
• Let the application of other RFI detection
  algorithms

15

• THANKS FOR YOUR ATTENTION