Low-CNR inverse synthetic aperture LADAR imaging demonstration with atmospheric turbulence 4/19/2016 Russell Trahan, Bijan Nemati, Hanying Zhou, Michael Shao, Inseob Hahn, William B. Schulze Presented by Russell Trahan
Testbed ○○○○○○ CNR Derivation ○○○ Experimental Data ○○○○○ Conclusion ○ Summary Goals: • Demonstrate ISAL functionality in photon-starved conditions. • Find a metric that can predict the success/failure of PGA based on the return signal strength. Outline: • Testbed hardware setup and data processing • Basic setup for low-CNR • Atmospheric turbulence synthesis • Data pipeline • CNR • CNR definition for a single range-bin (including detector noise) • Various metrics based on CNR • Image quality metric to compare to metrics based on CNR • Experimental Data • High CNR functionality tests • Low CNR imaging examples showing PGA failure at mean CNR=~0.25 4/19/2016 SPIE 9846-14 2
Testbed ○○○○○○ CNR Derivation ○○○ Experimental Data ○○○○○ Conclusion ○ Testbed Hardware Setup and Data Processing 4/19/2016 SPIE 9846-14 3
Testbed ●○○○○○ CNR Derivation ○○○ Experimental Data ○○○○○ Conclusion ○ Transceiver / Target Layout Target Range Cross Top View Range Side View Range Line Target Circle Target PZT Target Rotation Stage 4/19/2016 SPIE 9846-14 4
Testbed ●●○○○○ CNR Derivation ○○○ Experimental Data ○○○○○ Conclusion ○ Transceiver Assembly To Target Receiver Transmitter Local Oscillator 4/19/2016 SPIE 9846-14 5
Testbed ●●●○○○ CNR Derivation ○○○ Experimental Data ○○○○○ Conclusion ○ Transmitter Designs • No atmospheric turbulence • Fiber termination and collimating lens • Atmospheric turbulence 1. Fiber Termination 2. Collimating Lens – collimate light from fiber 3. Iris – truncate Gaussian beam to FWHM 4. Focusing Lens – focus collimated light through the phase wheel 5. Phase Wheel – introduce phase error 6. Speckle Image – focal point of focusing lens 7. Magnification Lens – magnify the speckle image onto the target 1 2 3 4 5 6 7 4/19/2016 SPIE 9846-14 6
Testbed ●●●●●○ CNR Derivation ○○○ Experimental Data ○○○○○ Conclusion ○ Testbed Overview Phase Wheel Mag. Lens LO Focus Lens Transmitter Receiver 4/19/2016 SPIE 9846-14 7
Testbed ●●●●●● CNR Derivation ○○○ Experimental Data ○○○○○ Conclusion ○ PGA Summary Our best results came from starting the window at 75% of the cross range extent, allowing 𝜒 to converge to nearly zero, then decreasing window size by 25%. Repeat until window is ~10 pixels in cross range. ˆ P Converged ? P k P P e i i i 1 m ˆ ˆ ˆ ˆ * detrend arg P P m m m m n , m 1, n m ' 1 n Over-sampling in range or including range- bins with very low CNR shouldn’t influence the phase increments. Simply includes noise in summation. 4/19/2016 SPIE 9846-14 8
Testbed ●●●●●● CNR Derivation ○○○ Experimental Data ○○○○○ Conclusion ○ CNR Derivation and Image Quality Metrics 4/19/2016 SPIE 9846-14 9
Testbed ●●●●●● CNR Derivation ● ○○ Experimental Data ○○○○○ Conclusion ○ CNR Definition Estimate of carrier strength • CNR is defined as StdDev of estimate of carrier strength • Measurement can be modeled as 2 2 N N exp i N N exp i N 0, N 0, d h L S d h L S s SN NEP • The carrier for a single range bin is N N exp i d h L S s • Shot noise variance is N 2 L SN d 2 2 • Detector NEP noise variance is P 2 NE NEP 2 2 2 h • Model is used to estimate the carrier strength and its variance N N N L S S CNR 2 2 4 2 N 1 4 N 4 4 var N N S S NEP NEP NEP L S 2 2 2 4 2 4 2 N N N d h d h d h L d h L d h L N N 1 S d h S CNR for N S 2 2 N 1 d h S 2 2 d h d h 1 N for N d h S S d h R. L. Lucke and L. J. Rickard, "Photon-limited synthetic-aperture imaging for planet surface studies planet surface studies," Applied Optics, vol. 41, no. 24, pp. 5084-5095, 2002. 4/19/2016 SPIE 9846-14 10
Testbed ●●●●●● CNR Derivation ●● ○ Experimental Data ○○○○○ Conclusion ○ Quality Metric Selection Quality metrics based on Quality metric based on pre-PGA data: post-PGA result: • # Photons in each range-bin • Image Contrast-to-Noise Ratio Maximum, Mean, Sum, Sum of squares mean foreground −mean background • 𝐷 = stdev background • CNR of mean photons per range-bin • Foreground region is determined based • CNR of each range-bin on a priori knowledge of the target. Maximum, Mean, Sum, Sum of squares • PGA performance cannot be assessed as • Phase progression Variance of each 𝐷 decreases past 1. range-bin Minimum, Mean, Sum, Sum of squares Primary Question: What quality metric has a consistent value at the threshold where PGA doesn’t work ? Immediate Question: What quality metric has a consistent value when the image contrast-to-noise ratio is 1? 4/19/2016 SPIE 9846-14 11
Testbed ●●●●●● CNR Derivation ●●● Experimental Data ○○○○○ Conclusion ○ Contrast depends on Cross-Range Extent Considering only a single range bin and a consistent CNR: • The image contrast is inversely proportional to the number of cross-range bins populated by the target. 𝑂−1 𝑄 𝑜 2 = 𝑂−1 𝑞 𝑙 2 1 Parseval’s Theorem: σ 𝑜=0 𝑂 σ 𝑙=0 • • Sum of a single range- bin’s magnitude over all pulses must equal the mean of the cross-range pixel values. FFT • If a single cross-range pixel is filled by the target, contrast will be high . • If several cross-range pixels are filled by the target, contrast will be low . *This idea is confirmed in the experimental data presented later. 4/19/2016 SPIE 9846-14 12
Testbed ●●●●●● CNR Derivation ●●● Experimental Data ○○○○○ Conclusion ○ Imaging Examples ~2m Range to Target Range Cross Top View Range Side View Range Line Target Circle Target 4/19/2016 SPIE 9846-14 13
Testbed ●●●●●● CNR Derivation ●●● Experimental Data ● ○○○○ Conclusion ○ Sample Low CNR Result Contrast: 1.9 average over many pulses # LO Photons per pulse: 5.05e+12 Top View # Range Bins: 33.9 # Photons per Range Bin: • Max: 1.92 • Mean: 0.55 • Sum: 18.54 • Sum of sqr: 18.52 CNR of Mean Photons per Range Bin: 0.27 CNR of Active Range Bins: • Max: 0.66 • Mean: 0.24 • Sum: 8.15 Difference • Sum of sqr: 3.14 4/19/2016 SPIE 9846-14 14
Testbed ●●●●●● CNR Derivation ●●● Experimental Data ●● ○○○ Conclusion ○ JPL Logo on Spectralon Chirp Rate 2THz/s Pulse Length 34 ms Acq Time 60 s Front View Mean CNR 2.76 4/19/2016 SPIE 9846-14 15
Testbed ●●●●●● CNR Derivation ●●● Experimental Data ●●● ○○ Conclusion ○ Satellite Image Chirp Rate 2THz/s Pulse Length 34 ms Acq Time 60 s Top View Mean CNR 4.5 Illumination Beam 4/19/2016 SPIE 9846-14 16
Testbed ●●●●●● CNR Derivation ●●● Experimental Data ●●●● ○ Conclusion ○ Line Area Contrast vs Mean CNRs Top View 4/19/2016 SPIE 9846-14 17
Testbed ●●●●●● CNR Derivation ●●● Experimental Data ●●●●● Conclusion ○ Line Area Lin Line Tar arget t (t (top row) Low Mean CNR Images Area ea Tar arget (bo (bottom row) Top View Contrast: 5.9 Contrast: 1.3 Contrast: 1.8 Mean CNR: 1.32 Mean CNR: 0.31 Mean CNR: 0.79 No Turbulence Turbulence No Turbulence Contrast: 3.2 Contrast: 0.84 Mean CNR: 1.07 Mean CNR: 0.31 No Turbulence No Turbulence 4/19/2016 SPIE 9846-14 18
Testbed ●●●●●● CNR Derivation ●●● Experimental Data ●●●●● Conclusion ● Conclusions • Testbed build to perform ISAL studies • Short 2m or long 400m range-to-target • Synthesized atmospheric turbulence • High and very low CNR capabilities • CNR Derivation • Rigorous derivation of CNR for a single range-bin • Quality metric for overall signal: “Mean CNR” • Quality metric for image: Contrast-to-Noise Ratio • Experimental Results • Target cross-range extent decreases image contrast (for constant CNR) • PGA can work for simple images down to ~0.25 CNR • Atmospheric turbulence raises minimum CNR threshold to ~0.75 4/19/2016 SPIE 9846-14 19
Sponsors 4/19/2016 SPIE 9846-14 20
Recommend
More recommend
