Skip to content

GOES HRIT / EMWIN

The GOES decoder establishes real 927 ksymbol/s HRIT/EMWIN reception from complex IQ. Validated CADUs bypass high-rate React event state and flow into a dedicated SATCOM product worker for virtual-channel, packet, file, and imagery construction.

Input and physical layer

Property Value
Current service target GOES-R HRIT/EMWIN near 1694.1 MHz
Decoder rate 1.854 MS/s, exactly 2 samples/symbol
Symbol rate 927 ksymbol/s
Modulation BPSK
Preferred bandwidth 1.4 MHz (1.2–1.854 MHz permitted)
ASM decoded 0x1ACFFC1D
FEC rate-1/2 K=7 Viterbi + 4-way interleaved RS(255,223)
Validated output 892-byte CADU

The documented SATCOM1 receive checkpoint uses a 2.4 MS/s capture; the common channelizer frequency-translates and resamples it to the manifest's 1.854 MS/s decoder rate. This is an example capture configuration, not a general hardware rate limit. Driver offset tuning can move the receiver's center artifact by changing physical LO/digital compensation; it does not change the operator's 1694.1 MHz VFO.

Configured frequency and target are not decoded spacecraft identity. Only synchronized, Viterbi/CCSDS Reed-Solomon-validated output establishes HRIT reception; relay identity and originating product source remain separate.

IQ-to-CADU pipeline

flowchart TB
    CU8[Receiver CU8 IQ] --> DC[Optional residual DC correction]
    DC --> CH[Frequency translate + resample to 1.854 MS/s]
    CH --> SRRC[31-tap SRRC matched filter]
    SRRC --> CARRIER[Carrier recovery]
    CARRIER --> TIMING[Mueller–Müller symbol timing]
    TIMING --> EASM[Encoded ASM correlation]
    EASM --> VIT[Soft K=7 rate-1/2 Viterbi]
    VIT --> NRZM[NRZ-M decode]
    NRZM --> ASM[0x1ACFFC1D frame sync]
    ASM --> RAND[CCSDS derandomization]
    RAND --> RS["4-way RS(255,223)"]
    RS --> CADU[892-byte validated CADU]

The coded frame is 16,384 bits plus the acquisition prelude; Viterbi output is 1,028 bytes before the CCSDS/RS stages. The Reed–Solomon code uses 32 roots with the implementation's CCSDS-compatible fcr=112, prim=11 geometry. Corrected symbol counts and uncorrectable frames are reported separately.

An uncorrectable FEC frame is not automatically a carrier-lock loss. RF Sentry tracks carrier/timing acquisition, ASM/frame sync, FEC outcome, source discontinuity, and product integrity as distinct stages.

CADU-to-product pipeline

flowchart TB
    CADU[Validated CADUs] --> VCDU[VCDU / virtual channel]
    VCDU --> MPDU[M_PDU parsing]
    MPDU --> CCSDS[CCSDS packet reconstruction]
    CCSDS --> LRIT[HRIT/LRIT headers and files]
    LRIT --> CLASS[Profile/product classification]
    CLASS --> SEG[Segmented image assembly]
    CLASS --> EMWIN[EMWIN text and graphics]
    CLASS --> DCS[DCS/environmental records]
    SEG --> ARCH[(IndexedDB metadata + OPFS files)]
    EMWIN --> ARCH
    DCS --> ARCH
    ARCH --> UI[SATCOM product workspaces]

Assemblies survive interleaving, duplicates, counter wrap, out-of-order segments, mid-file entry, and bounded reconnect gaps where the protocol permits. Assembly state is also isolated by lane, receiver/route, job, live/replay source and DVR source epoch. Within that context, source/channel/product/observation geometry groups segments; matching identifiers across sessions are not enough. Strips from two scans are never animation frames and must not be combined into one image. Entering mid-file cannot fabricate unavailable headers or pixels.

SatcomProductService keeps segment raster decoding, mosaic encodes, derived color and interactive presentation on separate imagery-worker queues. The product worker remains the file/archive coordinator; native image processing must not monopolize the receive decoder or move into React. See product ownership.

Original files and raw unknown products are retained. Complete, incomplete, corrupt, unsupported, and abandoned states are explicit. A valid CADU is not proof of a complete file or image.

Imagery and derived color

Supported Rice-compressed imagery is reconstructed at declared geometry. Natural-looking false color uses the documented goestools/wx-star C02+C13 recipe with exact-scan/source/sector/navigation matching and preserved provenance. Full-disk 5,424×5,424 inputs can produce a native 5,424×5,424 derivative; thumbnails and animation previews are separate reduced caches.

Day/Night Color adds a matching C07 observation and the C07/C13 broadcast toa_brightness_temperature lookup tables in kelvins. The implementation compares calibrated infrared temperatures, not arbitrary display grayscale, and calculates its day/night blend from navigation and observation time. Missing calibration or compatible inputs cannot be replaced by invented physical values. imageryColor.ts owns these recipes and input checks.

Natural-looking false color and Day/Night Color are approximations, not true color or exact GeoColor. Partial products retain coverage masks instead of borrowing cloud pixels from another scan. Source IDs, content hashes, times, recipe and render revisions accompany each derivative.

Native playback uses an operator-budgeted rolling prepared-frame cache and prefetches around the playhead. Viewer pan/zoom is a display transform, not a color regeneration. Compact navigation previews do not set native viewer or export resolution. Observed and estimated animation frames remain distinct; display refresh rate is not acquisition cadence. Original, qualitative enhancements and derived color remain distinctly labeled.

The main Imagery page applies an operator-configurable minimum coverage to automatic thumbnails and loops. The default is 90%; exact percentages and received/expected segment counts remain visible, and qualifying partials keep an explicit Partial label and their coverage mask. This policy never hides evidence from the detailed viewer, which retains every partial in its selected history window. Main and detailed loop windows/pacing are browser-local preferences restored after reload; play state and playhead are not restored. Loop windows are relative to the newest observation in the sequence, not wall clock time. Natural-looking and Day/Night preparation remains exact-scan and bounded/sequential, with live reconstruction retaining priority.

The upper workspace header groups source/server/stream identity with Configure source and consumption controls, while live appliance RF synchronization and Dock / close remain separate on the right. The compact HRIT/LRIT header shows the current operation, activity, scoped primary/historical progress and issues. One controller owns both presentations. Disclosures stay attached to their triggers; active activity precedes waiting work and terminal history. Earlier recorded incidents remain distinct from an active local recovery blockage.

The workspace chevron collapses the entire receiver/navigation rail. Its preference is separate from the tuner page. Decoder panel positions remain saved and the active content stays mounted while the rail is hidden; narrow windows use a drawer containing navigation and pinned-signal controls.

GOES source controls and status use compact, single-line items. Transfer opens current-operation details, pending/imported revision counts, downloaded bytes and elapsed time. Server opens upstream state, exact counters, source receipt time, publication/progress times and the historical range. A newly published revision can represent old backfill; the latest processed input does not establish complete historical coverage. Long explanations stay in these disclosures. Less essential inline fields move into details at narrow widths; the status/tab strips can scroll horizontally without widening the page.

Reception diagnostics show labelled time and value axes, matching series keys, latest values and mouse/touch/arrow-key sample inspection. The two-minute plots break at missing samples, resets and long reporting gaps; cumulative counters remain exact and separate from rates. Info controls provide hover explanations and keyboard/touch-accessible details. Appliance data still arrives at roughly 1 Hz: this UI update does not change appliance reporting or poll cached data faster. Direct-IQ alignment samples use their existing maximum 4 Hz while the diagnostics view is visible, scheduled in the pop-out's own window. Explicit reception tests keep their existing collection cadence when another view is selected. Plot storage stays bounded to 120 appliance or 480 direct-IQ samples.

For appliance/Server records, reconstruction automatically selects genuine memory64 where available and the compatible lowered-memory64 artifact otherwise. Both share C++ reconstruction, exact 64-bit identities and checkpoints. Configure source → Advanced / Diagnostics shows the selected backend without a warning for normal compatibility operation. Record-source operation explains storage prerequisites, memory limits and pending device acceptance. The same source dialog also provides an independent browser-local resource policy. Constrained admission coordinates reconstruction turns, preparation and cache residency while retaining native pixels and the full archive. Workstation keeps its existing parallelism and hot-cache targets. See the client resource policy for selection rules, overrides, accounting limits and safe recovery.

The imagery toolbar keeps its title, ordering, help and Recent previews … ready · … cached inline. Cached bytes describe compressed preview data, not total browser RAM or GPU use; the disclosure explains the accounting. Longer imagery explanations live in Help rather than an additional heading row.

Natural color and Day/Night preparation appears in the selected preview's upper right corner, with distinct checked/total and ready counts. Its disclosure states the window and queue scope. Full Disk uses the black surround where space permits; CONUS uses a compact backed overlay. These controls are separate display elements and are not written into exported image pixels.

Received GEOS navigation supplies a display-only Earth-limb mask for satellite thumbnails, retained previews and native playback. Outside-Earth pixels render black even when retained scalar/derived colors encode white space. No white-pixel replacement is used: valid clouds, measurements, legends and in-Earth missing-data patterns retain their values. Products without supported received geometry stay unmasked. Existing decoded images and native textures remain reusable; originals, exports and archive retention are unchanged.

Mesoscale footprints use subdued lavender strokes. The existing toggle applies to Full Disk and derived CONUS, using the displayed frame's observation time, received navigation and crop coordinates. Selecting a footprint still opens the corresponding received visit. An unset preference defaults off; saved preferences are preserved.

Replay and diagnostics

Diagnostics record requested and applicable gain/AGC, logical capture center, physical LO where known, VFO, residual correction, sample discontinuities, lock stages, validated CADU rate, RS corrections, uncorrectable frames, and product progress. dBFS is never labeled dBm.

The explicit alignment checks observe the existing lane; they must not retune, change gain/rate or stop another decoder. Reports preserve cancellation and partial results, counter resets, configuration changes and separate source, decoder-input and product-handoff loss accounting. Carrier/timing lock, frame sync, FEC outcome and product completeness are not interchangeable validation states. The detailed test/report contract remains in docs/verification/satcom/goes-hrit/report.md.

Recorded IQ can be replayed through correction off/on using identical samples. Product observation time remains separate from reception and processing time, so replayed weather cannot appear newly issued.

Tests and qualification

goesHritRawIqPipeline.test.ts exercises deterministic raw IQ through the production demodulator, acquisition, Viterbi, derandomization, RS, and CADU output, including noise, frequency/DC offsets, corruption, and discontinuity. satcomHritLrit.test.ts, archive, catalog, imagery, export, gallery, viewport, and native-playback tests cover the product path.

Qualification also compares independently sourced OTA/frame/image evidence against goestools/libaec-derived behavior; generated fixtures are not the sole reference. The current implementation supports the documented GOES-R profile. Other HRIT/LRIT services require a validated profile rather than being labeled supported because their names are similar.

Ledger and remaining proof

Phase 11A in docs/roadmap/aviation-satcom.md records the receive/alignment foundation and capture-center correction evidence; Phase 11B records the first receive-to-product candidate. Its live counters, mid-file imagery limits and remaining gates belong to that checkpoint, not a new measurement on the current appliance. Later code includes color treatments, geographic overlays and richer weather/environmental interpretation; the original open checklist is not evidence that all those features are absent today.

Generated raw-IQ fixtures, environment-gated independent recording/reference comparisons and operator live checks are separate evidence. Preserve extended reliability, reconnect/restoration, second-lane isolation, product UX and additional-profile qualification gates. No ledger checkbox or historical acceptance claim changes with this consolidation.