Sebastian Torres

1
NEXRAD Range-Velocity Ambiguity Mitigation

• Sebastian Torres

Fall 2004 Technical Interchange Meeting
2
Part One

• Recap of last TIM

3
Surprise! GMAP

• GMAP adopted as the ORDA GCF in FY04
• Ice et al. (2004) performed initial assessment
• Blackman window required for 50 dB suppression
• Input noise required for small number of samples
• Devoted good part of FY04 to study GMAP
• Implemented in MATLAB from source code

4
Yet Another Windows Update

• Window choice
• Sachidananda et al. (1998) recommended using von
  Hann window
• Blackman window is more aggressive
• Larger standard error of estimates (needs to be
  quantified)
• Recommendation Consider an adaptive scheme that
  uses the presence/strength of clutter for window
  selection

5
Lets Make Some Noise

• Noise estimation
• A rank order noise estimation algorithm is
  invoked if noise is not provided to GMAP
• Filter notch width depends on noise level
• A good noise estimate is critical for the
  filters performance
• SZ(8/64) phase coded out-of-trip echoes appear as
  noise
• Recommendation Provide reliable noise estimate
  to GMAP

6
Filling the Void

• Spectral reconstruction
• GMAP fills notched spectrum using Gaussian
  interpolation
• Biases are minimized
• Process assumes coherent signal spectrum
• Recommendation
• Clutter with strong signal use spectral
  reconstruction
• Clutter not with strong signal bypass spectral
  reconstruction (notch filter)
• GMAP modification enable/disable spectral
  reconstruction

7
Its just a phase

• Time-series reconstruction
• GMAP operates in the power spectrum domain
• Phase information is lost
• SZ-2 requires time-series for cohering process
• Recommendation Save unfiltered phase spectrum
  and use zeroes in the gap
• GMAP modification return number of spectral
  components with clutter

8
SZ-2 Clutter Filtering (I)

• Conditions for filtering
• Determined by maps
• Bypass map
• Clutter censor zones
• Determined by clutter strength
• Clutter power is not available in maps
• GMAP removed power is a good estimate of clutter
  power only for CSR gt 0 dB
• Recommendation
• Use maps as in legacy WSR-88D
• Use total power if clutter is not with the two
  strongest trips

9
SZ-2 Clutter Filtering (II)

• Sequence of operations
• Clutter must be removed first
• Cohere to trip with clutter
• Apply GMAP
• Censoring
• Recommendation Do not recover weak signal if
  clutter is not with the strong signal

10
Part Two

• Status of the SZ-2 Algorithm

11
SZ-2 Evolution

• Initial recommendation on Aug 15, 2003
• Interim recommendation on May, 2004
• Incorporation of GMAP (with a couple of
  modifications)
• Ability to handle clutter in any trip
• PNF optimization
• Spectrum width computation
• New recommendation on June 1, 2004
• GCF bypassing using CSR (in addition to maps)
• New censoring rules and refined thresholds
• Thresholds may require further refinement after
  operational tests
• Errata on June 22, 2004

12
Scan Strategy

• Long PRT (non phase coded) used to retrieve
• Filtered powers
• GMAP removed powers
• Spectrum widths
• Short PRT (phase coded) used to retrieve
• Strong and weak trip velocities
• Strong trip spectrum width
• Weak trip spectrum width from the long PRT scan

13
The SZ-2 Algorithm (I)

• Determine overlaid trips
• Determine clutter location(s) 3 cases
• Window time series
• If needed, filter clutter
• Cohere to trip with clutter
• Apply GMAP

14
The SZ-2 Algorithm (II)

• Determine strong and weak trips
• Cohere to trips with recoverable signals
• Compute autocorrelations
• Cohere to strong trip
• Compute strong trip velocity
• Apply PNF

15
The SZ-2 Algorithm (III)

• Cohere to weak trip
• Compute weak trip velocity
• Compute strong and weak trip powers
• Compute strong trip spectrum width
• Censor unrecoverable data
• Eight censoring rules
• Recommendation Censoring thresholds should be in
  adaptation data

16
SZ-2 Censoring

• Three types of returns
• Significant
• Overlaid-like (purple haze)
• Noise-like (not shown on displays)
• SZ-2 censoring occurs in
• Doppler velocities
• Spectrum widths

17
SZ-2 Censoring Rules (I)

• Noise-like cells
• (1) Low SNR in the long-PRT scan
• Cells with non-significant powers are not
  considered as candidates for recovery during the
  short-PRT scan
• KSNR is specified in the VCP definition
• (2) Low SNR in the short-PRT scan
• Takes care of advection between long- and
  short-PRT scans
• KSNR is specified in the VCP definition

18
SZ-2 Censoring Rules (II)

• Overlaid-like cells
• (3) Low SNR
• Out-of-trip signals appear as noise
• Thresholds for strong and weak trips are Ks and
  Kw
• (4) Weak trip not recoverable
• Recovery region from plots of SD(v2) in the S1/S2
  vs. sn1 plane
• Different regions for narrow and wide weak-trip
  spectrum widths

19
SZ-2 Censoring Rules (III)

• Overlaid-like cells (contd)
• (5) High CSR
• Strong clutter residue makes recovery of overlaid
  signals very difficult
• Thresholds for the strong and weak trips are
  KCSR1 and KCSR2
• (6) Clutter location
• Weak trip recovery only feasible is clutter is
  with strong trip

20
SZ-2 Censoring Rules (IV)

• Overlaid-like cells (contd)
• (7) Large weak trip spectrum widths
• Spectrum widths are derived from long-PRT
• sv saturates at 4.8 m/s with PRT 1
• Threshold is sv,max
• (8) Triple or quadruple overlay
• SZ-2 can recover at most two overlaid trips
• Third or fourth strongest trips are censored

21
Future Enhancements

• Proposed SZ-2 algorithm provides significant
  improvement compared to legacy algorithms
• Identified 4 areas for further improvements
• Use of GMAP without spectral reconstruction
• AP clutter suppression
• Recovery of overlaid echoes with comparable
  powers
• Weak trip spectrum width computation
• Need more research
• Cost-benefit analyses

22
Part Three

• Performance of the SZ-2 AlgorithmCase Examples

23
Set Up

• Data collected with KOUN radar
• RRDA with analog or digital receiver
• Experimental VCP, lowest elevation angles, 5
  scans at each elevation angle
• Non PC, long PRT
• Non PC, medium PRT
• PC, medium PRT
• Non PC, short PRT
• PC, short PRT
• Proposed SZ-2 algorithm (June 04) completely
  implemented in MATLAB

24
Stratiform Precipitation
25
Stratiform Precipitation
26
Stratiform Precipitation
27
Stratiform Precipitation
28
Stratiform Precipitation
29
Stratiform Precipitation
30
Stratiform Precipitation
31
Convective Precipitation
32
Convective Precipitation
33
Convective Precipitation
34
Convective Precipitation
35
Convective Precipitation
36
Squall Line
37
Squall Line
38
Squall Line
39
Squall Line
40
Squall Line
41
MCS-Squall Line
42
MCS-Squall Line
43
MCS-Squall Line
44
The End