2 nd SKADS Workshop 10-11 October 2007P. Colom & A.J. Boonstra 1 Outline -progress in RFI mitigation (methods inventory) -system design & RFI mitigation:

2nd SKADS Workshop 10-11 October 2007 P. Colom & A.J. Boonstra 1

Outline

- progress in RFI mitigation (methods inventory)

- system design & RFI mitigation: what and where

System design & RFI mitigationDS4T3 (OPAR, ASTRON, INAF-IRA, UORL, CSIRO)


Deliverable Achieved1 RFI mitigation methods inventory fall 2007

2 Influence on data quality Y3Q4

3 Impact of moving interference sources fall 2007

4 Cost effect. and tech. Requirements SKA site Y3Q4

5 Demonstrations with EMBRACE, BEST … Y4Q4

6 RFI mitigation strategies for the SKA Y4Q4

phased-arrays

System design & RFI mitigationDS4T3 (OPAR, ASTRON, INAF-IRA, UORL, CSIRO)


RFI Mitigation Methods InventoryReport, June 2007

• Introduction

• Spectral selectivity

• Temporal selectivity

• Spatial Selectivity

• Multi-dimensional techniques

• Implications for SKA and conclusions


Figure 1: (a) A classical filter bank implementation. (b) polyphase implementation of any filter. (c)polyphase implementation of a filter bank. (d) Example of filter banks (M=64). The input signal was the sumof an impulse, a sweeping sine wave and 2 switching sine waves: (d.1) h(k) is a 64 long rectangular impulseresponse. The filter bank is just a FFT. (d.2) h(k) is a 64 long Blackmann-Harris window. (d.3) h(k) is 640long low pass filter.

(a) (b)

(c)

(d.1) (d.2) (d.3)

Spectral selectivitypolyphase filterbank

ALMA memo 447 (J. Bunton) for cascaded PFB


Spectral selectivitynarrow band RFI elimination

Figure 4 : Analysis/synthesis process with quasi perfect reconstruction filter bank. Discrete Cosinustransforms have been implemented. Narrow band interferences are visible on the time-frequencyrepresentation (left). After analysis, polluted channels have been removed and the signal has beenreconstructed through a polyphase synthesis filter bank. The resulting time-frequency plane is given on theright


Temporal selectivityBlanking

• Detection theory based on hypothesis testing


Temporal selectivityBlanking

• Single-antenna detection: Pfa and PD are known

Pfa = Q(2PD = Q(2/ (1+INR))

• Multiple-antenna (p) detection (spatial-temporal) matched spatial detector: compare the received energy from the

interferer to the noise

test : data covariance matrix, combined with known interferer direction

Pfa : same PD = Q(2/ (1+p.INR))

residual after blanking : INRres 1 / p.N1/2

N: number of samples


Spatial selectivityFiltering

• Algorithms are based on modifications of data covariance matrix by a spatial filter, such that:

Pkak = 0 (ak direction of interferer)

Pk applied to covariance matrix: interferer energy nulled

• when ak is unknown ?

=> find eigenvalues and eigenvectors• a correction (matrix) has to be applied to the filtered

covariance matrix• Constraint: astronomical signal power << interferer power• residual after blanking :


Multi-dimensional techniquesCyclostationarity

cyclostationary process : statistics are periodic with time

Random binary signal: temporal view

covariance : time origin as random

covariance : time origin as constant


system design & RFI mitigation ASTRON/ISPO SSSM SKA monitoring results 2005/2006

virtual site, i.e. median of maxima of curves from the four sites visited : South Africa, China, Australia, Argentina

Cf. SKA monitoring protocol 2003 S.Ellingson et al (SKA memo)

Number of ADC bits:- 3 to 7 effective bits - depending on f, BW, site- if nonlinearities for short timescales are allowed: only 3 to 5 bits are needed


system design & RFI mitigationdata transport bottleneck

From RFI & data coding perspective: • use large subband bandwidth from stations to central site• break bands into more subbands / isolate bands with strong RFI• apply (fixed) spatial nulls at station level (“cheap”)• apply parametric techniques (more expensive; specific to coding

scheme)


system design & RFI mitigation effectiveness

Bottom line: one can mitigate RFI down to the level that it can be detected

So: delay RFI mitigation to the last stages in the datastream where data compression reduces the RFI mitigation SP load (beamforming, post correlation integration), unless…

• for dynamic range reasons- linearity requirements of LNAs after BF (PAF)- reduce number of bits (data transport reduction / digital stages PAF/AA)

• RFI is strongly spatially distributed- then local spatial filter makes more sense, at stations or between several stations

• RFI spectral bandwith does not match channel bandwidth- all methods

• RFI temporal characteristics - excision of s bursts close to antenna; drawback: loss of gain information


system design & RFI mitigation effectiveness

Stacking of methods is not usefull unless …...different domains are combined, e.g.

• RFI source subtraction, sidelobe cancelling and spatial filtering in

arrays are all spatial methods – in general not much use combining them

• parametric methods (Glonass/DVB suppr., Ellingson et al) and spatial filtering


system design & spatial filtering

DOA, subspace techniques: order Nant3

Rank-one subspace techniques, single source DOA: order Nant2

If direction known, and apply to beamformer or correlator: cheap!

DOA and subspace estimation usually is expensiveespecially if it needs to be done at a high update rate

Applying fixed filters, known fixed directions• Most fixed transmitters: easily ~20 dB supp. • If propagation modifies spatial structure:

add closely spaced nulls / increase subspace to be removed• Very cheap method if combined with beamformer of correlator Applying on-line varying filters, moving interferers• Both for fixed and moving transmitters: good suppression• Filter distorts uvw data as well, but can be restored under certain conditions • Expensive method (online matrix operations)• Drawback: affects the beamshape => hampers on-line calibration, “smoothness

criterium”


may be costlycan be used in combination with other methods

parametric techniques

(assuming wide bands)

may be costly; changing sidelobes may impair calibration

somewhat better suppression than fixed; tracking possibilities

varying spatial filters, sidelobe canceller

lower SP load at output station beamformers

-[excision]

fluctuating beam may impair calibration

reduce strong RFI enables the use of less ADC bits / lessens LNA req.

varying spatial filtering, including sidelobe canceller

Antenna beamformers (e.g. PAF)

difficult; needs careful calibrationreduce strong RFI enables the use of less ADC bits / lessens LNA req.

fixed spatial filtering

bookkeeping very costly; impairing gain estimate otherwise

low SP load unless booking is done on excised samples; fast transients

excision (assuming no subband filtering is done yet)

may be complex; may be time consuming

very flexible; can be added when necessary; relatively cheap

Spatial filtering,

parametric techniques, …

Post processing

-can be done at short timescales and short bandwidths; common practice

excisionCorrelation

influences UVW data points;

may impair calibration

may be applicable at shorter timescales than at location of correlator output

Interstation sidelobe cancelling/ spatial filtering, moving sources

Pre-correlation

more complex operation; connection wit cenral systems

very cheap; reduce data transport rate to central site

fixed spatial filterStation beamformers

Con’sPro’sMethodSignal path

Applicability in SKA

Documents

2 nd SKADS Workshop 10-11 October 2007P. Colom & A.J. Boonstra 1 Outline -progress in RFI mitigation (methods inventory) -system design & RFI mitigation: