Een telefoon die werkt
als een echt instrument.
Signal Collector legt doorlopend de sensor-, radio- en netwerkgegevens vast die je telefoon kan bereiken, rechtstreeks in de telefoon — zonder cloudopslag en zonder dat meetgegevens het apparaat verlaten.
Omvang van de MVP
Eén sessie.
De hele context.
- 01GPS, GNSS en sensoren van het toestel
- 02Mobiele netwerken: LTE / 5G / 3G / 2G
- 03Wi‑Fi, Bluetooth en netwerkcontext
- 04Doorlopende internetkartering via Wi‑Fi en mobiel
Weg uit de test,
de wereld in.
Versie 1.0.0 is de eerste publieke uitgave. Het instrument zelf verandert niet; nieuw is dat iedereen het kan installeren, een uur kan uitproberen voordat er betaald wordt, en de schermen in de eigen taal kan lezen.
Een uur meten, van het huis
Elke telefoon krijgt een uur meettijd per jaar, te verdelen over zoveel metingen als je wilt — het gereedschap is bedoeld om in het veld te proberen, niet op een schermafbeelding. De resterende tijd en de datum waarop hij terugkomt staan direct boven de Start-knop.
Gekocht, niet gehuurd
Ontgrendelen is één aankoop in de app, tegen de Google Play-prijs voor jouw land. Geen abonnement, geen vervaldatum, en het volgt het account naar elke andere telefoon waarop je inlogt.
Rechten uitgelegd voordat erom wordt gevraagd
De eerste start begint niet meer met een stapel systeemvensters. Een scherm vertelt waarvoor elk recht dient — en dat Android zendmasten en Wi-Fi aan locatie koppelt — en het systeem vraagt pas nadat je op de knop hebt gedrukt.
Schermen in jouw taal
De app en deze site spreken een hele reeks talen, gekozen door de telefoon of alleen voor Signal Collector ingesteld. De export volgt nooit mee: die blijft in SI, UTC en met Engelse kolomnamen, zodat twee bestanden uit twee landen één dataset blijven.
Ready for the world
and for the Store.
Version 0.10.0 completes the move to the production identity, adds full English and US units, closes the release blockers and gives the instrument a calmer, more precise visual system — without changing how it is operated.
English and Czech, neither one a bolt-on
Text is no longer produced inside the collection layer, and errors are no longer recognised by their wording. The language can be switched for the app alone, and an English phone never gets half-Czech screens. Numbers follow the locale; the export deliberately stays stable.
Metres, feet and miles
Units can follow the phone automatically, or be forced to metric or imperial in the settings. Only the screen and the notification are converted; the database and the scientific export always stay in SI.
A release with no silent traps
The production bundle requires the upload key, targets Android 16 / API 36 and carries the final identity app.signalcollector. Corrected permission checks close the crashes that followed a revoked permission, and the updated SSH client brings security fixes.
Around the globe
runs time itself.
Inside the globe sits only what has a bearing. Quantities without direction got their own bands outside the horizon, the operator composes the instrument themselves — and measured data can finally leave the phone by a route other than the share sheet.
Four quantities in the ring
Noise, temperature, pressure and now light. The angle along a band is time, not bearing — one minute split into seconds, so the head of the trace sweeps like a second hand with a gap behind it where values have aged out of the minute.
You compose the view
The middle button switches off whatever the cursor rests on. A switched-off row stays in the list with its value, only struck through: the filter changes the picture, not the measurement.
Data can leave on their own
The gear icon gained S3, WebDAV and SFTP destinations. Passwords are encrypted with a key from the Android Keystore, an address without https is rejected, and the server key fingerprint is verified.
You can see
where the signal came from.
The pulsing background circle gave way to a globe of the sky. Satellites sit where they actually are — and anything without a direction lights the whole ring instead of inventing a bearing.
Satellites in their real place
Android reports azimuth and elevation for GNSS only, and the app has recorded them since day one — there was simply nowhere to see them. Colour separates the constellation, size and brightness carry C/N0, a filled dot counts towards the fix. Verified on a Pixel 10a: 20 satellites from four constellations, 14 in the fix.
What has no bearing lights the whole horizon
Wi‑Fi gives strength, not azimuth; Bluetooth gives none either, and cells report only their identity. Samples without direction therefore ripple the entire ring rather than a sector — a sector would claim the signal came from there. The ripples also rotate, so not even the tallest of them points anywhere.
It tilts with the phone, not just turns
Lying face up it looks into the zenith, held upright it looks at the horizon — and what was near the horizon moves behind the phone. Collapsed, the card leaves a flat compass with a needle to north: calm, no flicker.
Scrolling that
never jumps away.
While hiding, the header used to take part of the gesture for itself and the content was re-measured under the finger. Now it only observes the scroll and takes nothing from it.
Content follows the finger the whole time
The finger used to drag while the page stood still until the header had collapsed — and on reversing direction it expanded again first. The header now sits above the content and takes no part in scrolling; it simply moves away, taking the status row with it.
Nothing is re-measured mid-scroll
It hides by offsetting at draw time, not by changing its height, so the content inset stays constant. Measured on a Pixel 10a: 0.30 % janky frames and no missed vsync across ten drags over the threshold where the header switches.
It returns the moment you pull down
Direction decides, not position — you do not have to scroll back to the top to get the header. Because it also returns mid-page, it got an opaque backdrop; otherwise the title would sit on top of the card text.
A scan that does not
pretend to be a measurement.
Android throttles Wi‑Fi scans and keeps handing back the same result meanwhile. The app used to write it down again and again — and the export then claimed a measurement every five seconds.
A repeated scan is recognisable
Every attempt records the state FRESH, THROTTLED or UNCHANGED together with the age of the result; per-AP records are written only for a fresh scan. Without that, the data could not tell you where measuring actually happened — and for a coverage map, that is the whole difference.
Six times denser sampling
Measured on a Pixel 10a: with throttling on, 70 rejected attempts against 45 fresh scans; with it off, zero rejections and a fresh scan every 4.2 s at the median. At walking pace that is one sample every 6 metres instead of every 40.
Points on a floor plan, throughput per network
GPS gives no position indoors, so a place marker was added — you stop, mark a point, walk on. And a new optional source visits the Wi‑Fi networks you list one by one and measures their actual throughput, not just their signal.
Every second
gets its own row.
Continuous internet measurement no longer leaves holes in the timeline. When measuring is impossible, the reason is written down — and a Wi‑Fi network that is gone is recognised at once, not ten minutes later.
A second without a measurement is still a row
A capacity burst or a manual speed test holds the link for several seconds, and the data used to be left with a gap — a full minute in the worst case. Now it says BUSY, naming who held the link. A missing second cannot be told apart from a failed measurement; a written one can.
A dead Wi‑Fi is recognised immediately
After losing Wi‑Fi, the app stayed bound to a network that no longer existed, and over six hundred samples claimed the internet was not answering. The handle is now dropped as soon as the system stops knowing about it, and the app waits for a new Wi‑Fi with internet.
A sample knows which Wi‑Fi it came from
SSID and BSSID were added to the quality record — until now only the direct Wi‑Fi source knew them, and measured quality could not be tied to a network. Android placeholder values (RSSI −127, frequency −1) are also left empty instead of looking like readings.
From your pocket you can tell
how the run is going.
The notification for a running measurement no longer just says “collecting data”. It shows the duration and the record count, refreshes every five seconds and speaks up when data stops being stored.
Time and count in the notification
On a dark screen you see 00:02:43 · 29,880 records instead of a terse line. The duration is pure collection time, the same as on the status card — a pause between Stop and Start does not count towards it.
Data loss reports itself
As soon as samples are dropped, their count joins the line, and the title changes when writing stalls or fails. It is the one thing worth reporting before the operator looks at the screen again.
Ready for Live Updates
The notification declares itself to the system as a Live Update — Android's counterpart to the iOS Dynamic Island, a strip inside the status bar itself. On the Android 16 build under test it does not render yet, because that part of the system is not switched on; the app is waiting for it, ready.
The colour tells you
what was just captured.
A wave crosses the background every time something is written — and now in a colour that says where the sample came from. From across the desk you can see not only that data is flowing, but what is being heard right now.
A colour per source
Green is sensors, blue networks, amber GNSS and violet Bluetooth. Sensors run at 50 ms, so green is a continuous swell with the other colours flashing through it — a blue flash is a network scan, a violet one Bluetooth.
The wave keeps off the text
It used to be drawn over the content as well, tinting the title in the header, the messages and the status row. Now it sits at the very bottom: it passes under the header but leaves the type alone.
Content runs under the bars
When the header hides on scroll, the cards continue under the status bar instead of stopping at it and leaving an empty strip above. The wave shows through under the bars too.
Start no longer overwrites
work in progress.
Start resumes the last measurement instead of quietly replacing it with a new one. Beyond that, the phone lets your pocket know when the connection drops or data stops being stored.
Start continues, it does not restart
Until now every Start opened a new measurement and the previous one became unreachable — it stayed in the phone, but only the new one could be exported. Now it continues, and a clean slate comes only from Reset, which deletes the previous measurement.
A pause does not count towards the duration
The duration is the sum of the time actually spent collecting, not the span from the first Start to the last Stop — for a run interrupted overnight that would claim fifteen hours of collection. In the export it is running_ms.
Interruptions are in the data
Every stop and every resume adds its own session record with a timestamp. The data therefore shows when collection ran and when it waited, so a pause is not mistaken for a hole in coverage.
A buzz at a hole in coverage
Losing the connection is a soft thud, regaining it two light taps, a failed write three firm knocks. The phone is in your pocket while you walk, and these are the events the screen cannot tell you about. Every buzz goes into the data as haptic, so a window can be cut around the vibration.
The counter runs live
The count is no longer sampled. At two thousand records per second it used to jump by more than five hundred; now it flows, and only the digits slow enough to read are animated. The figures have fixed slots and sit on the baseline.
A panel instead of a form
Start, Reset and Stop are solid colour fields — an unavailable switch turns translucent but keeps its colour. Export got its own plate with a label on which the size of the pending file grows, and a green wave crosses the background with every written sample.
How much ground
we actually covered.
This version adds distance walked next to the duration, in a way that does not credit metres to a phone lying on a desk. Beyond that, the capacity burst shows its progress and technology selection is driven by presets.
Distance walked, in metres
The status card shows Distance instead of a static source count. The value is stored with the session, so a finished measurement still shows it after an app restart, and in the export it is distance_m.
Metres, not GPS noise
A walker covers about 1.4 m between one-second fixes, while the uncertainty of a single fix is usually larger. Distance is therefore not measured between neighbours but by moving an anchor once it passes the threshold max(3 m, accuracy).
Location accuracy on show
The accuracy of the last fix sits under the metrics. When the position is too coarse to count towards distance, the card says so outright, so a number that will not move has an explanation.
The capacity burst shows its progress
A new record for the start of a burst puts its beginning into the data, not only its result. Meanwhile the UI runs a countdown next to the transport, and a burst interrupted by Stop is not left hanging.
Technology presets
“All” and “Clear all” became a single switch, with Default in front of it. The app recognises hand-tuned sensor values, asks before overwriting them and offers to keep them as the Custom preset.
Legible messages
A dark treatment tinted by message type, dismissible by dragging down. An error stays on screen for 15 seconds, an ordinary notice disappears after four.
Internet mapping
while moving through a building.
Continuous transfer mapping got its own card ahead of Start, a heartbeat began watching collection from the inside, and a code audit cleaned up the bindings to Android callbacks.
A card of its own for mapping
A separate card keeps continuous mapping apart from the one-off speed test. It can measure Wi‑Fi, cellular data or both transports at once over a specific Android Network.
A data estimate up front
The interface shows the upper bound before you start: at most 204 MB per 30 minutes for one transport and 407 MB for both.
A heartbeat over collection
A running check of the collectors, the writer and the write queue. The session then distinguishes an ordinary Stop, a writer failure and an interruption after the process died.
The export admits its gaps
The header carries the last stored timestamp, the trailing gap and the counts of accepted, written and dropped samples.
Audited callbacks
Registrations for GPS, GNSS, NMEA and navigation messages are visible in the export itself. Bluetooth Classic stores the states of the discovery cycle as well as the devices actually found.
Fixes from the code audit
The duration of a finished measurement is computed from stored timestamps, stopping the mapping leaves no NetworkCallback hanging, and sensors deliver data on their own thread.
Maintenance that
stops dropping samples.
No new measurement feature. A version born from a code audit, fixing the reliability of collection, writing and export, and moving the format to V3.
Dropped samples on show
When writing falls behind, the count appears in the app and goes into the header as dropped_samples. Samples used to vanish unannounced; ending a measurement no longer loses the last of them either.
Ranges hold
A setting outside the allowed range is pulled back into it, so a zero interval no longer spins a scan with no pause.
A disabled Bluetooth is recognised
With the adapter off, the sources are not marked as measured and ADAPTER_DISABLED goes into the export instead of a blank.
Collection off the main thread
The GNSS, location, Wi‑Fi RTT and cellular callbacks run aside, so raw GNSS no longer slows the controls. The microphone is released only after the read completes.
Format V3
SIGNAL_COLLECTOR_TXT_V3 adds the dropped_samples header and drops the source_timestamp_ns column, which only duplicated elapsed_realtime_ns.
Schema 2 without losses
The database moves by migration and existing measurements survive. Temporary files created for sharing are cleaned up, so the cache no longer grows by a whole measurement.
Transfer quality,
measured continuously.
Alongside the one-off speed test there is now a per-second stream of samples of real download, upload and RTT — bound to a specific network and enriched with radio context.
A sample every second
The measurement runs from Start to Stop and produces roughly one network_quality_sample per second.
What a sample holds
A 64 KiB download, an 8 KiB upload and three small HTTPS RTT probes. The output carries the real transfer times, short-term throughput, RTT min/avg/max, jitter, errors and timeouts.
Capacity burst
Every 30 seconds a network_capacity_probe is produced with a 1 MiB download and a 128 KiB upload.
Forced transport
The test can be sent over a specific Wi‑Fi or cellular Android Network, even when both are available at once.
Radio and location with the sample
Every sample carries the last Wi‑Fi or cellular radio context and the last known location, including its age.
Quality, not a record
The small samples are labelled as measurements of quality over time, not of maximum internet speed. The server layer supports the CLOUDFLARE, CUSTOM and LOCAL modes.
Controls that
stay where they are.
A small interface release. Actions no longer move with the state of the measurement, the top bar gives room to the content, and a message can be dismissed when it is in the way.
Download and Share stay put
The buttons keep their place. Without a finished measurement they are dimmed and inactive, not hidden.
A vanishing top bar
It hides when you scroll down and comes back when you scroll up.
Recommend from the bar
Prepares a new e‑mail with no recipient filled in, carrying a subject, a message and links to the app.
Dismissible messages
Status notices are a bottom message that can be closed; otherwise they disappear on their own after a longer interval.
The storage note moved down
The note about local data storage now sits at the very bottom, visually subdued.
Active scan
and a measured speed test.
This version introduces an experimental active cellular scan with an exact reason for refusal, and a full speed test bound to the transport actually chosen.
Active scan events
Unambiguous ACTIVE_NETWORK_SCAN_REQUESTED, STARTED, RESULT, COMPLETE and FAILED instead of guessing support from the phone brand.
The exact reason for refusal
A failed scan stores the numeric code, the exception type, the message, the carrier privileges state and the state of the MODIFY_PHONE_STATE permission.
Kept apart from passive cells
The result carries scan_mode=ACTIVE_NETWORK_SCAN and only the operator Android supplied; an unknown one stays UNKNOWN.
A test over the chosen transport
Manually over MOBILE, WIFI or sequentially BOTH, with the traffic genuinely bound to the requested Android Network. The options offered follow the validated networks available.
A complete test summary
speed_test_result carries download and upload in Mb/s, bytes and times, the minimum, mean, median, maximum and p95 RTT, jitter, packet loss, errors and the server — together with the last cell or Wi‑Fi context.
Collection that survives
The measurement continues in the background through a foreground service, preferences survive a restart, and from an estimated 10 MB upwards the export preselects GZIP.
Installatie
Uitsluitend te installeren
via Google Play.
- 1
Installeer vanuit Google Play — zonder uitnodiging, zonder testerslijst, zonder wachten.
- 2
Meet tot een uur per jaar zonder iets te betalen.
- 3
Ontgrendel onbeperkt meten vanuit de app zodra het uur niet meer volstaat.
Google Play is het enige distributiekanaal: het ondertekent de updateketen en nergens wordt een directe APK aangeboden. De app zelf heeft geen account van ons en geen server van ons nodig om te draaien.
Proef, prijs en ondersteuning
Een uur per jaar
om te beslissen.
Een veldinstrument laat zich niet beoordelen op schermafbeeldingen, dus geeft Signal Collector elke telefoon een uur meettijd per jaar — genoeg om een route te lopen, de ring te zien vollopen en de export te openen. Er wordt meettijd geteld, niet het aantal sessies, dus je kunt hem op meerdere plekken en meerdere dagen besteden. Is hij op, dan is ontgrendelen één aankoop in de app tegen de Google Play-prijs voor jouw land: geen abonnement, geen verlengdatum, en hij geldt op elke telefoon die met hetzelfde account is ingelogd.
Feedback loopt via de recensie in Google Play; elke nieuwe recensie wordt binnen een dag opgepakt, omgezet in een melding en beantwoord, en correcties reizen mee met de volgende update. De betaalde app bevat geen reclame, geen analytics en geen SDK van derden binnen de meting — de prijs is wat in plaats daarvan de ontwikkeling betaalt.
Wat de app werkelijk verzamelt
Van fysieke sensoren
tot de netwerken om je heen.
Elke bron is voor een meting afzonderlijk aan of uit te zetten, en bij geselecteerde technologieën is ook het interval in te stellen.
Mobiel netwerk
Bedienende en naburige cellen, LTE/NR/3G/2G-parameters, registratiewisselingen en de experimentele actieve scan.
GNSS en locatie
1 Hz GPS, satellieten, ruwe GNSS-metingen, klok, AGC, navigatieberichten, NMEA en antenne-informatie.
Wi‑Fi en RTT
De huidige verbinding, access points in de buurt, Wi‑Fi 7 MLO, mogelijkheden en RTT/FTM-afstanden tot ondersteunde access points.
Bluetooth
Klassieke detectie en de BLE-advertentiepayload, inclusief fabrikant- en servicegegevens.
Sensoren en geluid
Beweging, magnetisch veld, druk, licht, nabijheid, rotatievector en een indicatief dBFS-niveau zonder opgeslagen audio.
Netwerk en overdracht
Monsters per seconde van echte download, upload, RTT, jitter en time-outs, capaciteitsbursts van 30 seconden, en gelijktijdige binding aan een specifiek Wi‑Fi- en mobiel netwerk.
SIM en eSIM
Slot, provider, MCC/MNC, roaming en welke kaart de data draagt — cellen op zichzelf benoemen de zender, niet wie er keek. Noch ICCID noch EID wordt vastgelegd: dat zijn permanente abonnementsnummers die zelfs een telefoonwissel overleven.
Veldtest, 10 augustus 2026
Gegevens van een echte
telefoon.
Alles hieronder is gemeten op een Google Pixel 10a met Android 16. In versie 0.7.2 verzamelde één meting in drie stukken 87.597 records: de duur groeide van 2:49 via 7:21 naar 8:12, het meetnummer veranderde nooit, en meer dan twintig minuten pauze tussen de stukken telde niet mee in de duur. De afstandscijfers komen uit een sessie van negen minuten met versie 0.7.1 op dezelfde telefoon — het afstandsfilter is sindsdien niet veranderd.
De eerste 274 fixes hadden een nauwkeurigheid tussen 9 en 114 m. De afstanden tussen naburige fixes optellen zou 429 m geven, of 28 m nadat grove fixes zijn weggefilterd; het anker voorbij de drempel verplaatsen gaf nul. De testtelefoon had geen SIM, dus de mobiele tak van de kartering meldde het netwerk terecht als niet beschikbaar.
Versies en materiaal
Installatie, wijzigingen
en broncode.
De repository blijft privé. Ingelogde medewerkers vinden er de broncode, getagde releases en APK-checksums.
Privacy
Meetgegevens verlaten de telefoon nooit.
De app werkt offline en bevat geen SDK voor analytics, reclame of crashrapportage — niets dat op de achtergrond over je zou kunnen rapporteren, want er zit simpelweg niets in dat dat zou kunnen. De optionele doorlopende kartering van de overdrachtskwaliteit gebruikt de publieke endpoints van Cloudflare Speedtest voor kleine HTTPS-testoverdrachten en periodieke bursts; het is geen meting van de maximale internetsnelheid. De module staat standaard uit. Koop je de ontgrendeling, dan verlaat één aankoopmelding de telefoon, zonder enige meetgegevens; die staat beschreven in het privacybeleid. Een export kan gevoelige locatiegegevens en identificatoren van nabije netwerken bevatten.
Hoe het werkt
Van een sensor
tot een regel in een bestand.
Elke bron loopt op zijn eigen tempo. Een monster krijgt een tijd en de laatst bekende locatie mee, gaat door een niet-blokkerende wachtrij en wordt in batches in de lokale database opgeslagen. Onderstreepte sleutels in het schema leiden naar het gegevenswoordenboek.
Instellingen van de verzameling
Dezelfde parameters die de app voor elke bron aanbiedt voordat een meting begint. De bereiken en de dataschatting komen overeen met de implementatie; een wijziging is meteen zichtbaar in het schema hierboven.
Bereik 10–1000 ms
Bereik 100–60000 ms
Bereik 5–300 s
Bereik 5–60 s
Bereik 5–300 s
InternetkarteringNetwerk voor de doorlopende meting
Met beide geselecteerd draaien twee aparte Android- Network -handles tegelijk.
Bereik 500–5000 ms
Bereik 1–1024 KiB
Bereik 1–256 KiB
Bereik 10–300 s
Bereik 64–10240 KiB
Bereik 8–2048 KiB
De schatting is een bovengrens: op een trage verbinding duurt een monster langer, er passen er dus minder in dezelfde tijd en er worden minder data overgedragen. Dit is geen meting van de maximale snelheid — waarom.