SAR · Satellite Imagery
A study log filling in first principles beneath an InSAR pipeline I built in production first
At Lumir I built the Sentinel-1 InSAR pipeline (DInSAR · SBAS · PSI) in production first. The pipeline ran fine, but first principles — why complex numbers, why a phase difference becomes displacement, what coherence really is — were missing. I'm filling that gap from the theory blocks up, wiring it back into hands-on experience.
Conversational learning with AI (ChatGPT · Claude) as a thinking partner, compiled into the Brain Trinity wiki. Raw conversations are preserved as-is; concepts are distilled into wiki pages my future self — and an LLM via RAG — can reuse.
When I studied what — dated by when each wiki page was created or each lecture recorded. Tap an active day to see that day's topics.
Foundations — the physics of signal & image
The "why" beneath the pipeline — from complex signals to TOPS scanning.
An SLC pixel is A·e^jθ. Euler's formula turns rotation into a single multiplication — that's why complex numbers — and the interferogram M×S* (conjugate product) extracts the phase difference automatically.
Move a 12.3 m antenna, observe the same target hundreds of times, phase-correct and sum (matched filter) — a huge virtual antenna emerges. That's what azimuth compression is.
Bandwidth sets range resolution (ΔR=c/2B); the synthetic aperture sets azimuth. Sentinel-1 IW's asymmetric 5 m × 20 m pixel — and why multilook is 3×1 — both follow from here.
Transmit long while sweeping frequency, compress short with a matched filter — pulse compression wins resolution and SNR at once, and it is the range compression in RAW→SLC.
Speckle — interference noise from scatterers; coherence — similarity of two SLCs (0–1); and the multilook trade-off that buys coherence at the cost of resolution.
TOPSAR sweeps the beam electronically to cover a 250 km swath — the IW1/2/3 subswath structure, paid for with 20 m azimuth resolution.
Operations & data products
The Sentinel-1 product system — what you actually download to start.
InSAR deformation
From phase difference to millimetre displacement — the theory of the production pipeline.
Δφ=(4π/λ)ΔR — one fringe is 2.8 cm in C-band. The line-of-sight constraint is resolved by combining ascending and descending passes.
Invert a network of tens-to-hundreds of short-baseline pairs into a mm/yr velocity field and time series. A reference point pinned to a stable site suppresses atmospheric (APS) noise.
Select only persistent scatterers on urban structures for full-resolution per-point time series — the inverse of SBAS. Proven end-to-end on 108,690 points in downtown Jukjeon.
SqueeSAR extends into vegetation and mountains with distributed scatterers PSI can't see — and the next map prepares for L-band NISAR with better canopy penetration.
Already running in production
This isn't desk theory — it's foundation work under a pipeline that's already in operation.
Built and operate the 5-tool pipeline across SNAP · ISCE2 · MintPy
DInSAR→SBAS time series end-to-end — through reference-point correction and QA
Urban PS time series completed, with phase-linking optimized from 5 h to 6 min