Real-time RFI Mitigation for the Upgraded GMRT Kaushal D. Buch Digital Backend Group, Giant Metrewave Radio Telescope, NCRA-TIFR, Pune, India kdbuch@gmrt.ncra.tifr.res.in
The (Upgraded) GMRT � Giant Metrewave RadioTelescope (GMRT) � Array consisting of thirty 45m diameter parabolic reflector antennas � A highly sensitive radio receiver system for observing astrophysical phenomena at low radio frequencies � Upgraded GMRT (uGMRT) � Upgraded GMRT (uGMRT) � Nearly seamless frequency coverage from 50 to 1450 MHz � Instantaneous receiver bandwidth from 32 MHz to 400 MHz � Increased bandwidth and receiver sensitivity to encounter increasing levels of man- made radio frequency interference (RFI) uGMRT 250-500 MHz and L-band released for observations on shared-risk basis http://gmrt.ncra.tifr.res.in/gmrt_hpage/Users/doc/GMRT-specs.pdf CASPER-2017, Caltech, USA 8/27/2017 2
Major Sources of RFI at GMRT 11 kV transformers around S-arm of GMRT array � Broadband RFI � Sparking on power-lines � Corona Discharge � Automobile sparking � Narrowband RFI � Communication transmitters � Broadcast TV / Radio � Satellites Data Courtesy: Pravin Raybole, RFI Group, GMRT CASPER-2017, Caltech, USA 8/27/2017 3
RFI at GMRT RFI Broadband RFI is stronger at lower radio frequencies Broadband RFI Narrowband RFI Examples showing impulsive time and frequency domain RFI observed at the GMRT RFI CASPER-2017, Caltech, USA 4
Need for Real-time Excision � Temporally impulsive RFI: Energy spreads post-FFT hence excision is needed before FFT. � Power-line RFI: Low duty cycle but high spectral occupancy � RFI is correlated in closely spaced antennas � Spectrally impulsive RFI: RFI excision useful for low time � Spectrally impulsive RFI: RFI excision useful for low time occupancy � Best possible time resolution: reduction in loss of astronomical data due to flagging (trade-off) � Leads to improvement in the receiver sensitivity CASPER-2017, Caltech, USA 8/27/2017 5
Location of Real-time Excision blocks FFT: Fast Fourier Transform MAC: Multiply & Accumulate IA: Incoherent Array PA: Phased Array RFI excision at multiple locations in the receiver chain (This talk will focus on real-time RFI for the uGMRT correlator) CASPER-2017, Caltech, USA 8/27/2017 6
GMRT Wideband Digital Backend (GWB) GWB: Reddy et al., JAI 2017 Real-time broadband RFI Mitigation is implemented on ROACH-1 FPGA board CASPER-2017, Caltech, USA 8/27/2017 7
Robust Detection using MAD • RFI in astronomical data – outliers make Gaussian distribution heavy-tailed (Fridman, 2008) • Excision (Baan, 2001, 2010) assumes that RFI is much stronger than the astronomical signal Robust threshold using Median Absolute Deviation for RFI detection • (Fridman, 2008) σ MAD = 1.4826(med(|x(i) – med(x)|)) (Fridman, 2008) σ MAD = 1.4826(med(|x(i) – med(x)|)) Impulsive RFI detection and filtering (Buch et al., JAI, 2016) • • Excision (filtering) by replacing the RFI affected samples by constant value or noise or threshold Robust threshold: median ± n* σ MAD Implemented in temporal and spectral domains for the uGMRT (Buch et al., RFI- • 2016, IEEE, 2016) CASPER-2017, Caltech, USA 8/27/2017 8
Broadband RFI Excision FFT: Fast Fourier Transform MAC: Multiply & Accumulate IA: Incoherent Array PA: Phased Array Broadband RFI filtering is carried out in real-time on Nyquist-sampled digital time- series (for each antenna) at 800 MHz CASPER-2017, Caltech, USA 8/27/2017 9
Estimation – Detection - Filtering Filtering can be bypassed CASPER-2017, Caltech, USA 8/27/2017 10
MAD-based RFI Detection and Filtering • Computation of real-time Median using the histogram method Available filtering options: 1. Clipping • Non-linear filtering Replacement with med(x+y) ≠ med(x) + med(y) 2. Constant value 3. Digital noise (Buch et al., • Values outside the range of JAI, 2014) [median ± n* σ MAD ] are treated 4. Median as RFI • RFI Detection is followed by filtering CASPER-2017, Caltech, USA 8/27/2017 11
Real-time Implementation MAD = med(|x(i) – med(x)|) Resource utilization: (8-bit input) 16k window MAD: 20% slices 4k window MoM: 18% slices � Uses the histogram method for median computation – explores parallelism on FPGA � Window size is 2T R /T S ) samples where T R is the (worst case) duration of RFI and T S is the sampling interval Long-lasting RFI: � Hold MAD values from consecutive windows in a memory buffer and compute the median (M) i.e. median of MAD (MoM) values (M m ) M m =M(MAD 1 ,MAD 2 ,...,MAD n ) Buch et al., RFI-2016, IEEE, 2016 12 https://casper.berkeley.edu/wiki/Impulsive_RFI_Excision:_CASPER_Library_Block
Real-time RFI Filtering: Features Filter one or both the polarizations First Median (External input) Includes option to bypass the filter Two main parameters – threshold value (N) and filtering options CASPER-2017, Caltech, USA 8/27/2017 13
RFI Count � Broadband RFI filter counts the RFI samples in a given time period along with the total number of samples � 64-bit counters � Two control signals – reset and hold � Manual hold : Asynchronous mode, Automatic hold: Synchronous mode � Manual hold : Asynchronous mode, Automatic hold: Synchronous mode � Flagging fraction = No. of RFI samples / Total samples � Timestamp along with the count for each antenna � Takes about half a minute to read data from all the 16 ROACH boards Works even if the filtering is OFF (bypass mode) CASPER-2017, Caltech, USA 8/27/2017 14
Test Procedure a. RFI emulator (controlled testing) b. Antenna signals c. System-level tests � Outputs: raw voltage (1.25 ns), beam mode (1.31 ms) and correlator mode (671 ms) mode (671 ms) � Simultaneous comparison between unfiltered and filtered data (different threshold and replacement options), beam and correlator outputs � Improvement characterized by the mean/rms ratio � Preliminary tests carried out for imaging radio sources with and without broadband RFI excision CASPER-2017, Caltech, USA 8/27/2017 15
Simultaneous Testing – Unfiltered and Filtered Digital copy of one antenna fed to two or four inputs for simultaneous comparison CASPER-2017, Caltech, USA 8/27/2017 16
Test Results – Raw voltage Comparison between MAD and MoM on unfiltered and filtered raw voltage Pulse input provided through RFI emulator, on-period 32 μ s total-period 240 μ s CASPER-2017, Caltech, USA 8/27/2017 17
Test Results – MAD vs MoM Total samples in 30 min. duration = 1440000000000 Samples flagged as RFI = 6146025673 (MAD)(0.43%) and 6270107796 (MoM)(0.44%) CASPER-2017, Caltech, USA 8/27/2017 18
Test Results – MAD vs MoM I (dB) = 10log(M F /M U ) M F , M U : running mean/rms for filtered, unfiltered beam output CASPER-2017, Caltech, USA 8/27/2017 19
Test Results – Beam and Correlator Simultaneous recording of beam mode (1.3 ms) and correlator mode (671 ms) with different thresholds replacement by zeros. CASPER-2017, Caltech, USA 8/27/2017 20
First Image Image Courtesy: DharamVir Lal Filtered RMS: 1.29 mJy Unfiltered RMS: 2.26 mJy 250-500 MHz, uGMRT, 16 antennas, Pol.-1 (left, without filter) & Pol.-2 (right, with broadband RFI filter), a factor of two (approx.) improvement in noise RMS CASPER-2017, Caltech, USA 8/27/2017 21
Spectral Domain RFI Filtering FFT: Fast Fourier Transform MAC: Multiply & Accumulate IA: Incoherent Array PA: Phased Array Real-time Narrowband RFI filtering is carried out at 0.671s integration on multi-core CPUs CASPER-2017, Caltech, USA 8/27/2017 22
Narrowband RFI Excision Two approaches to Spectral domain MAD filtering – 1. MAD-based filtering of each channel • over time (MFAT) and 2. MAD-based filtering across the spectral channels (MFAC) Estimation and filtering across channel is more suitable for real time applications – with • additional correction required for across the band gain variations. Current system: Normalization of spectral power is followed by MFAC • 8/27/2017 23
Narrowband RFI – uGMRT (250-500 MHz) CASPER-2017, Caltech, USA 8/27/2017 24
Summary � Broadband RFI filtering is available for the uGMRT user community � Long-term tests and imaging experiments show improvement in the performance of the system � Narrowband real-time RFI excision will be released soon! � Fine-tuning of broadband and narrowband RFI filtering techniques is in progress CASPER-2017, Caltech, USA 8/27/2017 25
Acknowledgements Imaging tests and data analysis Current members DharamVir Lal Kishor Naik Ruta Kale Sanjay Kudale Sanjay Kudale Ajithkumar B. Ishwara Chandra C.H. Viral Parekh (RRI) Viral Parekh (RRI) Yashwant Gupta Yashwant Gupta Past members GMRT groups Swapnil Nalawade GMRT Backend Team Control Room and Operations Shruti Bhatporia Short-term Engineering Interns 8/27/2017 26
Thank You! For queries, suggestions or comments, email: kdbuch@gmrt.ncra.tifr.res.in CASPER-2017, Caltech, USA 8/27/2017 27
Recommend
More recommend
