# Sky: Passive Aerial Early-Warning Node

Written by [Taki Sadik](../../about) · Last updated 2026-10-22 · 31 min read

A receive-only passive radar node built on a KrakenSDR five-channel coherent SDR: it listens to reflections of existing FM and UHF TV broadcasts bouncing off aircraft, computes range-Doppler plots with pyAPRiL, and scores every detection against an RTL-SDR ADS-B ground-truth receiver. The radar front end is receive-only: it emits nothing of its own.

## Overview

Canada's Arctic Over-the-Horizon Radar (A-OTHR), based on Australia's JORN
        design, adapted for RCAF/NORAD, with initial capability due December 2029, will provide
        long-range early-warning coverage of the northern approaches as one layer in a sensor picture
        that also includes conventional radar, airborne, and space-based sensors. Its companion
        program, Crossbow, is a classified Canada-US non-radar sensor network. You
        can't copy Crossbow, and you shouldn't try.

What you can build is the open, hobby-scale, receive-only analogue: a
        passive radar node. Instead of transmitting its own pulses, passive radar
        listens to reflections of existing broadcast signals, FM radio (88-108 MHz) and UHF TV
        (~470-600 MHz), bouncing off aircraft. Because the radar front end only receives, it needs no
        transmit licence, radiates nothing of its own to give away a position, and is cheap. The
        export-control and intercept-law checks that still apply are covered in the notes at the bottom. A KrakenSDR gives me five phase-coherent receive
        channels, which is enough to point one antenna at the broadcast transmitter (the
        reference channel) and the rest at the sky (the surveillance
        channels), then cross-correlate the two to find echoes.

This is Phase 1 of my [Canadian Defence
        Hardware Program](defence-hardware-program). It is the flagship sky-domain node: every detection it makes will
        eventually publish over MQTT into the multi-domain common operating picture, fused with the
        water and ground nodes.

## Proof of Build

Everything in the photo is already purchased and running. The ADS-B side went live first:
        a live aircraft map fed by an RTL-SDR, because a passive radar without independent ground
        truth is just a screensaver. The KrakenSDR and its antennas are the radar side; the laptop is
        where both halves meet for scoring.

What each part of the photo does:

- KrakenSDR: five receive channels sampled from one clock, so the phase between them is meaningful. One channel listens to the broadcast transmitter, the other four listen to the sky.

- Five whip antennas: bench antennas for testing the chain end to end. The directional reference and surveillance antennas in the bill of materials replace them outdoors.

- Raspberry Pi: runs the KrakenSDR software that captures the channels and computes the cross-correlation.

- Laptop, left plots: the range-Doppler waterfalls, where a moving aircraft shows up as a streak at its bistatic range and Doppler shift.

- Laptop, right map: tar1090, the ADS-B aircraft map. Every radar detection is checked against it before it counts.

## Bill of Materials

| Part | Role | Details | Est. US$ |
| --- | --- | --- | --- |
| KrakenSDR | Coherent radar receiver | 5-channel coherent receiver, 24 MHz to 1766 MHz, 2.56 MHz max channel bandwidth (2.4 MS/s is the stable operating sample rate) | 749 (price listed on the KrakenRF product page; confirm before ordering) |
| Antennas | Direction-finding array | KrakenRF 5-antenna set for testing | 199 |
| Directional antennas | Reference and surveillance channels | 2 Yagi / log-periodic for FM and UHF TV reference/surveillance | 80-160 |
| Compute | Processing and truth logging | Mini PC (Intel N100 or i5, 16 GB) running Ubuntu | 150-250 |
| ADS-B truth | Independent ground truth | RTL-SDR Blog V4, 1090 MHz antenna, coax | 50-80 |
| RF accessories | Connections and front end | SMA adapters, low-loss coax, optional FM/UHF filter or LNA | 60-100 |
| NanoVNA (optional) | Antenna tuning | Vector network analyzer for tuning the Yagis | 50-70 |

Total lands around US$1,350-1,600 at the current KrakenSDR price. I started with just the ADS-B kit
        (~US$100) so the ground-truth logging was collecting data weeks before the KrakenSDR
        processing chain existed.

## RF & Signal Chain

How the boxes actually connect on the bench. The two SDRs are deliberately on separate USB
        controllers, one saturated USB bus dropping samples is enough to ruin a coherent capture
        (see [Troubleshooting](#troubleshooting)).

| From | To | Cable / interface | Notes |
| --- | --- | --- | --- |
| FM Yagi (pointed at transmitter) | KrakenSDR CH1 (reference) | LMR-240 coax, SMA | Clean direct-path signal; keep this run short |
| Surveillance antennas (log-periodic / whips) | KrakenSDR CH2-CH5 | Matched-length coax, SMA | Cable lengths matched so phase calibration stays valid |
| KrakenSDR | Mini PC | USB 3.0 data + USB-C 5 V power | Dedicated USB port; quality power supply matters for coherence |
| 1090 MHz antenna | RTL-SDR Blog V4 | Coax, SMA | Mounted as high and clear as the room allows |
| RTL-SDR Blog V4 | Mini PC | USB 2.0 | Separate USB bus from the KrakenSDR |
| Mini PC | LAN / MQTT broker | Gigabit Ethernet | chrony NTP sync so radar and ADS-B timestamps agree |
| Mini PC | Laptop (monitoring) | SSH / HTTP over LAN | tar1090 map and range-Doppler plots viewed here |

Every cable on the receive side; nothing on this node ever transmits.

## Architecture

direct path

illuminates

reflected echo

FM / UHF TVbroadcast transmitter

Aircraft(reflects signal)

Reference Yagi(direct signal)

Surveillance antennas(sky-facing)

KrakenSDR5 coherent channels

Mini PCrange-Doppler + pyAPRiL

RTL-SDR ADS-Breadsb + tar1090

Scorerdetections vs truth

MQTT busfused detections

Figure 1: the transmitter is someone else's; the node only receives. Reference and surveillance channels cross-correlate into range-Doppler detections, which are scored against independent ADS-B truth before publishing to the fusion bus.

## How Passive Radar Processing Works

A passive radar never ranges by time of flight of its own pulse; it measures how much
          longer the aircraft's echo path is than the direct path. The reference Yagi
          records the broadcast waveform straight from the transmitter; the surveillance antennas record
          the sky, which contains the same waveform delayed by the extra transmitter-to-aircraft-to-receiver
          distance, plus shifted in frequency by the aircraft's Doppler. Cross-correlating the two
          recordings over a bank of Doppler shifts produces the range-Doppler map: every aircraft shows
          up as a bright peak at its bistatic range (the extra path length) and its Doppler.

The loudest thing on that map is never an aircraft: it is the direct path and
          the ground clutter, both sitting on the zero-Doppler ridge. Extended clutter cancellation (ECA,
          which pyAPRiL provides alongside its ECA-B and ECA-S variants) estimates and subtracts that
          ridge before detection, which is why the pipeline can see a small echo next to a mountain of
          clutter. Range resolution follows the bandwidth: the theoretical figure for passive radar is
          c / B, excess path length (the c / 2B figure is the monostatic range
          convention), about 1500 m for a 200 kHz FM station and about 50 m for a 6 MHz ATSC TV
          channel, which is exactly why the build compares both illuminators (T3 vs T4).

Figure: why the build compares illuminators: bandwidth sets the best-case bistatic range resolution (the speed of light over the bandwidth, c / B, the passive-radar excess-path convention). TV's 6 MHz channel resolves about 50 m where FM manages about 1500 m; away from the transmitter-receiver baseline the bistatic angle degrades this further, and real performance follows from the T3/T4 scoring runs.

Coherent IQ capture(KrakenSDR, calibrated)

Channel preparation(reference + surveillance)

Clutter cancellation(ECA / ECA-B / ECA-S)

Cross-ambiguity(range-Doppler map)

CA-CFAR detection(adaptive threshold)

Direction of arrival(pyArgus beamforming)

Scored hits(vs ADS-B truth)

Figure: the offline pipeline the build follows, mirroring pyAPRiL's modules: prepare the channels, cancel the clutter ridge, correlate into a range-Doppler map, detect adaptively, estimate direction, then score against the ADS-B log.

## How I Built It

**Q: 01 ADS-B ground truth first (~US$100)**

A: Before any radar work, I put up the truth receiver: an RTL-SDR Blog V4 with a 1090 MHz antenna, decoding Mode S / ADS-B with readsb (the maintained fork of dump1090) and serving a live map with tar1090. The decoder writes aircraft state, ICAO24, position, altitude, heading, speed, to /run/readsb/aircraft.json every second, and tar1090 archives rolling history under /run/tar1090/. This log is the scoring key for everything that follows: for any timestamp I can say exactly which cooperative aircraft were actually overhead and where.

**Q: 02 Illuminator survey**

A: Passive radar is only as good as its illuminators of opportunity, so before mounting anything directional I mapped my local broadcast landscape with rtl_power: which FM stations (88-108 MHz) and UHF TV transmitters (~470-600 MHz, ATSC) are receivable here, at what power, and on which exact frequencies. FM offers long range and strong signals; 8-VSB/ATSC TV has more bandwidth and therefore better range resolution. I cross-checked the strongest peaks against public transmitter databases to get each station's location, the bistatic geometry later depends on knowing precisely where the transmitter is.

**Q: 03 Antenna tuning with the NanoVNA**

A: A passive radar lives on faint echoes, so every dB lost to a mistuned antenna hurts twice (once on the reference, once on the echo). I swept the FM Yagi and the UHF log-periodic with the NanoVNA from the bench using NanoVNASaver, checking SWR across each target band and trimming the Yagi's driven element until the dip sat inside the FM band rather than at its edge. Target: SWR under 2:1 across the whole illuminator band, ideally under 1.5:1 at the chosen station's frequency.

**Q: 04 KrakenSDR bring-up**

A: The KrakenSDR is five RTL-SDR-class receivers in one box sharing one clock, covering 24-1766 MHz, the shared clock is what keeps all channels phase-aligned, which is the whole point for passive radar and direction finding. I run the open-source krakensdr_doa stack (HeIMDALL DAQ firmware plus the DSP/web layer) on the mini PC; a Raspberry Pi 4/5 works too, but the x86 box has the headroom for the processing chain in later steps.

**Q: 05 Phase calibration, and keeping it valid**

A: Five receivers sharing a clock are frequency-coherent, but each channel still has its own phase and amplitude offset from mixer and path differences. The KrakenSDR's built-in noise source injects a common wideband signal into all five channels, and the software measures and nulls those offsets. The critical operational rule: frequency and gain changes trigger Kraken recalibration, and any retune or sample-rate change invalidates the phase calibration, so recalibrate on every retune or sample-rate change and it must be re-run. I log the calibration timestamp alongside every capture so I can prove later that a capture was taken in a calibrated state.

**Q: 06 Reference and surveillance antenna geometry**

A: Two antenna roles carry the radar geometry. The reference Yagi points at the broadcast transmitter and captures the clean direct signal on channel 1. The surveillance antennas point at the area of interest on channels 2-5 and capture that same signal delayed and Doppler-shifted by aircraft reflections. Spacing between surveillance elements matters for direction-of-arrival work: the rule of thumb is half-wavelength spacing to avoid grating lobes, which is band-dependent, at FM frequencies the elements sit far apart, at UHF TV they pack tightly.

**Q: 07 Coherent IQ capture**

A: Processing development needs repeatable data, so before tuning anything live I built a recorder that pulls synchronized IQ frames from the HeIMDALL DAQ's network IQ interface (the same output the KrakenSDR mobile app consumes) and writes per-channel interleaved float32 files plus a small JSON sidecar with centre frequency, sample rate, gain, and calibration timestamp. The DAQ speaks framed MCHQ packets on that interface (magic, big-endian header with per-channel frequency and gain, then interleaved uint8 IQ), so the real recorder parses frames and converts to float32; the socket snippet below is schematic. A 10-30 second capture at 2.4 MHz sample rate is a few hundred MB across five channels, big enough to develop against, small enough to keep a library of.

**Q: 08 Range-Doppler processing with pyAPRiL**

A: The core DSP runs on pyAPRiL, the open-source Python passive-radar library on PyPI (pip install pyapril). KrakenRF's own ready-to-use passive radar code was withdrawn for regulatory reasons, so this chain is built on pyAPRiL instead of any Kraken-distributed implementation. The chain per coherent processing interval (CPI): prepare the reference and surveillance channels, suppress the direct-path breakthrough and stationary clutter with an adaptive filter (the ECA family, Extensive Cancellation Algorithm), compute the cross-ambiguity surface (surveillance correlated against reference over a bank of Doppler shifts), and run CA-CFAR detection over the resulting range-Doppler map. Aircraft appear as peaks at a bistatic range (extra path length vs the direct signal) and a Doppler speed. With multiple surveillance channels calibrated, pyAPRiL also estimates direction of arrival, which turns a blip into a bearing.

**Q: 09 Scoring harness against ADS-B truth**

A: A radar page without numbers is marketing, so the scoring harness is a first-class part of the build. For each logged ADS-B aircraft at each capture timestamp, I compute where it should appear in bistatic range-Doppler space, the extra path length via the known transmitter location, and the Doppler from its velocity projected onto the transmitter-aircraft and aircraft-receiver legs, then check whether the radar produced a CFAR hit there, and whether it produced hits where no aircraft existed. That yields the three numbers that define the radar: detection rate, false-alarm rate, and effective range per illuminator. FM and UHF TV get scored separately, because the honest answer is usually "it depends on the transmitter."

**Q: 10 Automation, logging, and multi-node fusion**

A: A sensor node you have to babysit is a demo, not a node. The DAQ, recorder, and processing chain run as a systemd service that starts on boot and restarts on failure; captures and score summaries rotate to the archive disk nightly. The endgame is 2-3 nodes at different sites publishing detections over MQTT, the same signed-detections bus the rest of the hardware program uses, into the multi-domain common operating picture, where intersecting range-Doppler-DoA solutions from multiple nodes become actual positions. Store-and-forward applies here too: nodes buffer through dead links and sync on reconnect, the DDIL pattern the rest of the program is built around.

## What CA-CFAR Does

After clutter cancellation, the range-Doppler map still has lumps: residual clutter, noise,
          and the aircraft peaks hiding in it. A fixed threshold cannot survive that
          landscape: set it low and every clutter lump is a false alarm; set it high and the quiet
          aircraft vanish. Cell-averaging constant false alarm rate (CA-CFAR, the detector pyAPRiL's
          hit processor provides) sets a different threshold for every cell of the map. For the cell
          under test it averages the energy of neighboring training cells, skips the
          immediately adjacent guard cells (so a real target's own energy does not
          pollute its threshold), multiplies by a scale factor chosen for the target false-alarm
          probability, and declares a hit only if the test cell exceeds it. The false-alarm rate stays
          roughly constant even as clutter and noise drift, which is exactly what the T5 overnight
          soak measures.

## Testing & Metrics

The protocol is built around the ADS-B ground truth, and every run is logged with its
        configuration (illuminator, gain, filter order, calibration timestamp) so results are
        reproducible rather than anecdotal. Each numbered run below is a repeatable procedure; runs
        get re-executed after any antenna, gain, or filter change.

### Test runs

- T1, ADS-B truth completeness. Setup: readsb/tar1090 running for a full week. Procedure: count daily decoded aircraft and check the history files for gaps. Metric: log uptime and messages/second. This has to be solid first, every later score inherits its errors.

- T2, Calibration hold time. Setup: KrakenSDR calibrated, then left streaming untouched. Procedure: re-check residual phase spread between channels at increasing intervals without recalibrating. Metric: time until range-Doppler quality measurably degrades. Output: the re-calibration interval the node enforces in software.

- T3, FM illuminator detection run. Setup: reference Yagi on the chosen FM station, surveillance antennas on the approach corridor, fresh calibration. Procedure: record captures across several busy traffic windows, process with the fixed pipeline, score against the truth log. Metric: detection rate per range band, and false alarms per hour.

- T4, UHF TV illuminator comparison. Setup: identical to T3 but on an ATSC transmitter. Procedure and metrics identical, the point is a per-illuminator comparison, because bandwidth and transmitter geometry differ.

- T5, False-alarm soak. Setup: pipeline running with CFAR at its working threshold during a quiet-traffic window (overnight). Procedure: count hits with no matching ADS-B track. Metric: false alarms per hour; this is what clutter-filter tuning is judged on.

- T6, Multi-node dry run. Setup: planned second node on the same illuminator. Procedure: both nodes record simultaneously, fuse detections over MQTT, and compare the intersected position with ADS-B truth. Metric: position error vs single-node geometry. Status: pending the second node.

### Targets, Measured Evidence, and Current Status

Evidence and targets are kept separate: the "measured to date" column holds only what has
        actually been recorded, and the target column holds what the build is aiming at.

| Metric | Target | Measured to date | Current status |
| --- | --- | --- | --- |
| ADS-B truth log uptime | > 95% over any scoring week | Running; uptime tracked, scoring weeks being selected | Running; uptime tracked, scoring weeks being selected |
| Detection rate (FM illuminator) | > 70% of ADS-B tracks in the primary coverage band | Not yet measured | In progress, T3 runs underway, not yet scored |
| Detection rate (UHF TV illuminator) | Comparable to FM within its coverage band | Not yet measured | Not yet measured, T4 pending |
| False-alarm rate | < 10 per hour at the working CFAR threshold | Not yet measured | In progress, clutter filter still being tuned |
| Effective bistatic range | Establish per illuminator, then publish | Not yet measured | Not yet measured, follows from T3/T4 |
| Calibration hold time | Quantify, then automate re-calibration on that interval | Not yet measured | In progress, T2 data being collected |
| Multi-node position error | Meaningfully better than single-node geometry | Not yet measured | Not started, second node not yet built |

Results are in progress and not yet published. The ADS-B truth log has been
        accumulating since the receiver went live; the range-Doppler chain is producing plots on the
        bench, and the scored detection/false-alarm dataset will be published once I have enough
        scored hours per illuminator to make the numbers meaningful. I am deliberately not quoting
        range or detection-rate figures until then.

## Legal & Safety Notes

## Troubleshooting & Lessons

- USB dropouts ruining coherent captures. Symptom: streaks across the waterfall and correlation that dies mid-CPI. Cause: both SDRs streaming through one overloaded USB controller, and an undervolted KrakenSDR. Fix: separate USB buses for the KrakenSDR and the RTL-SDR, a quality 5 V supply, short good cables, and dmesg -w watching for xHCI errors during every capture. Any capture with a dropout gets deleted, not processed.

- Range-Doppler map is all clutter ridge, no aircraft. The direct-path signal leaking into the surveillance channels was stronger than the echoes by orders of magnitude. Fixes, in the order that worked: physically separate and null the antennas (reference pattern null toward the surveillance array), then raise the ECA filter order in the processing chain, then confirm the calibration was current. If the zero-Doppler ridge still dominates, the geometry, not the code, is the problem.

- Coherence "randomly" lost after working an hour ago. Not random: frequency and gain changes trigger Kraken recalibration, and any retune or sample-rate change invalidates phase calibration. Fix: treat calibration as part of the run configuration, re-run it on every change, and log its timestamp with every capture so stale-calibration data can be filtered out of scoring.

- A nearby powerhouse FM station saturating the front end. A very strong local transmitter raised the noise floor across the band and produced intermod ghosts. Fix: back the gain down (coherence matters more than gain), pick an illuminator that is strong-but-distant, and if needed put an FM band-stop filter on the surveillance channels only, never on the reference.

- Detections that don't line up with ADS-B truth. First suspect was the algorithm; the actual cause was clock disagreement between the radar captures and the truth log, a few seconds of drift moves a fast aircraft well outside the tolerance window. Fix: chrony on the mini PC disciplining both pipelines off NTP, and a scoring-harness check that rejects runs where the truth log has gaps.

## Status & Roadmap

- Done: full BOM purchased and on the bench; ADS-B ground-truth receiver (readsb + tar1090) live with continuous position logging; FM and UHF TV illuminator survey with rtl_power; Yagi and log-periodic tuned with the NanoVNA; KrakenSDR streaming coherent IQ with the calibration discipline (recalibrate on every retune) in place; capture recorder writing CF32 plus metadata sidecars.

- In progress: pyAPRiL range-Doppler chain with ECA clutter filtering tuned against recorded captures; the T2 calibration-hold and T3 FM detection test runs; the scoring harness that aligns CFAR hits with the ADS-B truth log.

- Next: finish the T3-T5 scoring runs and publish the per-illuminator dataset (detection rate, false-alarm rate, effective range); systemd-managed always-on node with MQTT publishing; second node and the T6 multi-node fusion dry run into the common operating picture; export-control review with Global Affairs Canada before any radar code is published.

## Related Builds

- Multi-Domain Common Operating Picture, where this node's detections end up: fused with the water and ground nodes over MQTT with a human deciding.

- Passive Acoustic Drone Detector, the other receive-only sky sensor in the program, listening for rotor harmonics instead of RF reflections.

- Arctic Domain Awareness Digital Twin, the same fusion problem over open ADS-B, AIS, and satellite data at Arctic scale.

- Canadian Defence Hardware Program, the program hub with all builds, phased BOMs, and Canada-specific notes.

## Build Narrative: Phase by Phase

Phase 1: truth before radar. I built the ADS-B ground-truth receiver first: an RTL-SDR Blog V4 with a 1090 MHz antenna running readsb and tar1090, logging aircraft positions continuously. A passive radar without independent truth is unverifiable, so the truth pipeline had to exist weeks before any radar processing ran.

Phase 2: illuminator survey. Using rtl_power I swept the FM band (88-108 MHz) and the UHF TV band (~470-600 MHz) from the bench, cataloguing candidate illuminators by signal strength and geometry. This is also where the local FM powerhouse showed up, the one strong enough to saturate the front end, which became a design input rather than a surprise.

Phase 3: antennas and geometry. I built the FM Yagi aimed at the chosen transmitter as the reference channel, with log-periodic and whip surveillance antennas on matched-length coax, all tuned with the NanoVNA. Physical separation and pattern nulling were used to keep the direct-path signal from overwhelming the echoes in the surveillance channels.

Phase 4: coherent capture discipline. The KrakenSDR went onto the Ubuntu mini PC with the RTL-SDR deliberately on a separate USB bus, powered by a quality 5 V supply. Calibration is treated as part of the run configuration and re-run on every retune, since frequency and gain changes trigger Kraken recalibration. The capture recorder writes CF32 IQ plus metadata sidecars, with chrony NTP keeping radar and truth timestamps in agreement.

Phase 5: processing and scoring. The pyAPRiL chain cross-correlates reference and surveillance channels into range-Doppler maps, an ECA filter suppresses the zero-Doppler clutter ridge, and a CA-CFAR detector proposes hits. The scoring harness then aligns those hits against the ADS-B truth log inside a tolerance window, producing measured detection and false-alarm rates per illuminator.

## What You Need

Everything below mirrors the bill of materials on this page. The single non-obvious requirement is patience with calibration: the whole build assumes you re-calibrate on every configuration change.

| Category | Item | Notes |
| --- | --- | --- |
| Hardware | KrakenSDR | 5-channel coherent SDR, 5x R820T2 tuners, 24 MHz to 1766 MHz, 2.56 MHz channel bandwidth, ~11 W at 5 V |
| Hardware | Reference + surveillance antennas | FM Yagi for the reference channel, log-periodic/whip surveillance antennas, matched-length coax, NanoVNA for tuning |
| Hardware | Mini PC (Ubuntu) | Intel N100/i5 class with 16 GB RAM; keep the two SDRs on separate USB controllers |
| Hardware | RTL-SDR Blog V4 + 1090 MHz antenna | The ADS-B ground-truth receiver; its data is what makes every radar result scoreable |
| Software | KrakenSDR DAQ/DSP + pyAPRiL | Open-source KrakenRF software for capture, pyAPRiL for range-Doppler processing |
| Software | readsb + tar1090, rtl_power | Aircraft truth map and spectrum survey tooling, both free |
| Software | chrony, Python/numpy/scipy | NTP time discipline for truth alignment, numerical stack for the chain |
| Skills | RF fundamentals, Linux, Python DSP | dB arithmetic, gain staging, reading waterfall plots, coherent vs non-coherent captures |
| Accounts | None for receive-only | No accounts or licenses needed; export-control review with Global Affairs Canada is required before publishing radar processing code |

## Challenges and Lessons Learned

- USB dropouts ruined coherent captures. Streaked waterfalls and dying correlation traced to both SDRs sharing one overloaded USB controller plus an undervolted KrakenSDR. Lesson: give streaming SDRs separate buses and watch dmesg -w for xHCI errors during every capture; any capture with a dropout gets deleted, not processed.

- The zero-Doppler clutter ridge buried everything. Direct-path leakage into the surveillance channels was orders of magnitude stronger than echoes. Lesson: fix geometry first (antenna separation and nulling), then raise the ECA filter order, then confirm calibration. If the ridge still dominates, the geometry, not the code, is the problem.

- Coherence died on every retune. Frequency and gain changes trigger Kraken recalibration, and any retune or sample-rate change invalidates phase calibration. Lesson: calibration is part of the run configuration, and every capture gets its calibration timestamp so stale-calibration data can be filtered out of scoring.

- Clock drift masqueraded as algorithm error. Detections that did not line up with ADS-B truth turned out to be chrony/NTP disagreement between pipelines, a few seconds of drift moves a fast aircraft outside the tolerance window. Lesson: discipline time before debugging math.

## What Is Next

- Finish the T3 to T5 scoring runs and publish the per-illuminator dataset: detection rate, false-alarm rate, and effective range, all scored against ADS-B truth (measured results only, no targets dressed as data).

- Systemd-managed always-on node publishing scored detections over MQTT into the common operating picture.

- A second KrakenSDR node and the T6 multi-node fusion dry run, testing whether two receive-only nodes see the same aircraft consistently.

- Export-control review with Global Affairs Canada before any radar processing code is published, and only then.

- Evaluate ATSC illuminators for finer range resolution at higher frequencies, weighed against their weaker power density compared with FM.

## Frequently Asked Questions

**Q: How do I fix 'USB dropouts ruining coherent captures.'?**

A: Symptom: streaks across the waterfall and correlation that dies mid-CPI. Cause: both SDRs streaming through one overloaded USB controller, and an undervolted KrakenSDR. Fix: separate USB buses for the KrakenSDR and the RTL-SDR, a quality 5 V supply, short good cables, and dmesg -w watching for xHCI errors during every capture. Any capture with a dropout gets deleted, not processed.

**Q: How do I fix 'Range-Doppler map is all clutter ridge, no aircraft.'?**

A: The direct-path signal leaking into the surveillance channels was stronger than the echoes by orders of magnitude. Fixes, in the order that worked: physically separate and null the antennas (reference pattern null toward the surveillance array), then raise the ECA filter order in the processing chain, then confirm the calibration was current. If the zero-Doppler ridge still dominates, the geometry, not the code, is the problem.

**Q: How do I fix 'Coherence "randomly" lost after working an hour ago.'?**

A: Not random: frequency and gain changes trigger Kraken recalibration, and any retune or sample-rate change invalidates phase calibration. Fix: treat calibration as part of the run configuration, re-run it on every change, and log its timestamp with every capture so stale-calibration data can be filtered out of scoring.

**Q: How do I fix 'A nearby powerhouse FM station saturating the front end.'?**

A: A very strong local transmitter raised the noise floor across the band and produced intermod ghosts. Fix: back the gain down (coherence matters more than gain), pick an illuminator that is strong-but-distant, and if needed put an FM band-stop filter on the surveillance channels only, never on the reference.

**Q: How do I fix 'Detections that don't line up with ADS-B truth.'?**

A: First suspect was the algorithm; the actual cause was clock disagreement between the radar captures and the truth log, a few seconds of drift moves a fast aircraft well outside the tolerance window. Fix: chrony on the mini PC disciplining both pipelines off NTP, and a scoring-harness check that rejects runs where the truth log has gaps.

## Cite This Page

**Q: Citation formats**

A: Plain text
