Every megabit a vehicle camera sends over 4G is metered, billed and capped. The consumption follows from arithmetic a fleet can do on paper: the bitrate of the stream, the hours it runs, the events that trigger it. Three spenders sit on the meter, the watched minutes, the alarm parcels, the background telemetry, each behind a setting somebody controls; a budget’s fate is decided there, long before the bill arrives.
A vehicle video system spends data in a way no office system prepares a buyer for. The cameras record all day, yet almost none of that recording travels; what crosses the mobile network is a chosen slice, pulled on demand or pushed by an alarm. The bill a fleet pays tracks those slices. Understanding the consumption means understanding which slices exist, what each one costs per minute, then who decides when they run. The slices have names, prices and owners, three columns a fleet can fill in for itself.
Every number used is a working figure, produced by arithmetic from configured bitrates, recomputable on paper. Field-measured monthly totals across live fleets are their own subject, shaped by habits and seasons; the mechanics, the per-hour and per-event costs, decide what any measured total can mean.
A vehicle terminal encodes each camera twice. The main stream carries the full picture, high resolution at a high bitrate, written to the storage card as the legal record. The sub stream carries a lighter copy of the same view, smaller and cheaper, built for the mobile link. The split exists because the two jobs pull in opposite directions: the recording wants every detail the sensor can give, the network wants the smallest stream a viewer can still use.
Almost everything that travels over 4G in normal service is sub stream. A platform operator calling up a live view receives the light copy; the heavy copy stays on the card in the vehicle. One real terminal manual puts the configurable spread plainly: 720p channels adjustable from 768 kilobits per second up to about 6 megabits, 1080p from about 4 megabits up to the same ceiling, with sub streams sitting far below at a few hundred kilobits. The exact menus differ by maker; the two-tier shape is the industry’s standing answer to the cost of mobile data.
The platform chooses the tier at the moment of asking. The live-stream request the protocol defines carries a stream-type field alongside the channel number, so each viewing session names its price as it opens. The terminal answers with whichever copy was requested, which moves the cost decision off the vehicle and into the control room.
Every budget decision downstream starts from that price difference. A sub stream at 512 kilobits per second costs roughly a quarter of a gigabyte per viewing hour. A main stream at 4 megabits costs nearly two gigabytes for the same hour. Whoever chooses which stream answers a request is choosing, eight times over, what the request costs.
What the viewer loses at the light tier is sharpness, never the event. A supervisory question, where is the vehicle, is the bay clear, did the load shift, reads fine at a few hundred kilobits; reading a plate or a face is recording-tier work, done from the card after the fact. Fleets that internalise this stop paying main-stream prices for sub-stream questions. The habit also survives audits: a control room that can say which tier answers which question has already documented its data policy.
One conversion rule covers the whole subject: a stream of 1 megabit per second, held open for an hour, moves 450 megabytes, which is 0.45 gigabytes. The rule is nothing deeper than units, eight bits to the byte and 3,600 seconds to the hour, so it holds across makers, plans and stream types alike. Scale it linearly: half a megabit costs around 0.22 gigabytes an hour, 2 megabits about 0.9, 4 megabits about 1.8.
The rule turns any configuration page into a price list, the habit that carries this whole subject. The same arithmetic audits backward as readily as it plans forward. Forward: a depot manager who keeps one sub-stream channel open through an eight-hour shift at 512 kilobits has spent about 1.8 gigabytes, the same data a single hour of one full-rate main stream would have taken; four sub-stream channels watched together for that shift cost around 7 gigabytes, which is why a wall of live tiles left running is a budget decision even at the light tier. Backward: a SIM that came back from the carrier at 31 gigabytes for the month is making a claim the arithmetic can check, since 31 gigabytes at sub-stream rates is roughly 130 viewing hours, a number someone either recognises from the control room’s habits or cannot explain; the gap between those two reactions is where misconfiguration and abuse get found. The same two-way use covers events: a 20-second alarm clip pushed at 2 megabits is about 5 megabytes, so a vehicle that raised ten alarms a day with a clip and two photographs each has spent somewhere near 60 megabytes, call it under 2 gigabytes a month, a figure that looks tiny beside one careless afternoon of full-rate live viewing. One full pass through these comparisons on paper fixes the consumption hierarchy in the mind: continuous anything is expensive, event-driven anything is cheap, the resolution tier multiplies everything, the hours multiply everything else. The same sheet prices a policy before anyone signs it: a holiday weekend of standing watch on twenty coaches, two channels each at sub-stream rates for ten hours a day, comes out near 90 gigabytes a day, call it 180 across a two-day weekend, a number a manager can approve or refuse in advance rather than discover on an invoice. Per-channel thinking is the trap to avoid in these sums, since the costs that hurt are per vehicle times fleet; a setting that adds an innocent tenth of a gigabyte per vehicle per day walks out of a 300-vehicle depot at nearly a terabyte a month. A fleet that prices its own habits this way before signing a data contract knows whether it needs two gigabytes per vehicle or twenty, can price a control-room policy change before making it, recognises a wrong bill on sight. None of that knowledge needs a meter or a tool, only the one conversion and the willingness to multiply.
Precision beyond this is false comfort. Real streams breathe with the scene, since encoders spend fewer bits on an empty parked view and more on rain and motion; protocol overhead adds a sliver on top; a constant-bitrate setting holds the line flatter than a variable one. The working figures land within sensible distance of the bill, close enough to plan a contract on, close enough to spot a tenfold anomaly at a glance. What they cannot do is replace the carrier’s own count, which is the number the contract settles on.
Billing granularity adds its own sliver. Carriers meter in rounded blocks per session or per day, kilobytes up to megabytes depending on the plan, so a fleet of short frequent sessions pays slightly above the raw byte count. The effect stays in the low percents; it explains the last small gap between the paper sheet and the invoice, the gap that is nobody’s fault.
Storage runs on the same rule, one tier up. The main stream writing to the card at 4 megabits lays down about 43 gigabytes a day per channel, which is why card sizing is its own planning exercise and why nobody proposes pushing the full recording over 4G. The network tier exists because the arithmetic at the recording tier forbids anything else.
A buyer can hold the whole section in three sentences. Bits per second times seconds gives bits; divide by eight for bytes. An hour of 1 megabit is 0.45 gigabytes. Everything else is multiplication.
Live viewing is the consumption a fleet controls minute by minute, because every viewing minute exists on demand. The terminal sends nothing visual until the platform asks; the request opens a stream at a stated tier; closing the window closes the spend. The protocol builds this economy into the request itself: the platform names the channel and the stream type, so each viewing session is a deliberate, priced act.
The habits that govern it live in the control room. An operator who opens a vehicle for thirty seconds, confirms what needed confirming, then closes it, spends a few megabytes. A supervisor who tiles eight vehicles on a second monitor at the morning shift and forgets them spends gigabytes by lunch. The terminal obeys both. No setting on the device separates the two uses; separating them is a policy job for the control room.
Multiple watchers do not multiply the vehicle’s bill the way intuition expects. Platforms commonly take one upstream from the terminal and fan it out to every logged-in viewer, so the SIM pays once, no matter how many screens subscribe. The multiplication that does land on the SIM is channels: four channels opened on one vehicle is four streams up the same link, four times the spend.
Sampling covers the routine assurance job at a fraction of the standing cost. A control room that opens each vehicle for one scheduled minute per shift buys a daily visual confirmation for under five megabytes a vehicle, the spend a standing tile reaches inside its second minute. Fleets formalise the sample, a rota of spot-checks with a log line each, then reserve continuous watch for the runs that genuinely justify it.
Dispatch-style watching changes the calculus. Operations that genuinely need standing eyes on vehicles, a hazardous-goods run through a city, a coach fleet on a holiday weekend, are buying continuous sub-stream hours knowingly; the arithmetic prices that choice plainly: one channel watched around the clock for a month at 512 kilobits is roughly 160 gigabytes. A fleet that needs that should plan for it, price it as a named operating cost, then revisit the need each season; a fleet that merely drifted into it is funding a habit nobody chose.
Remote playback needs its own caution. Pulling stored footage back over 4G moves recording-tier data across the viewing-tier link; a half hour of main-stream playback costs near a gigabyte. The protocol supports asking for a lower-rate copy of stored footage for exactly this reason; control rooms that review long stretches remotely, at full rate, by habit, are running one of the least visible budget leaks the system has.
Alarm traffic is small per event, relentless in aggregate, the slice of the budget least suited to squeezing. When a safety alarm fires, the terminal pushes the evidence parcel the regulations shaped: a short clip from the recorder’s buffer, photographs from the named cameras, the data record that ties them to the alarm number. A 20-second clip at 2 megabits is about 5 megabytes; the stills add fractions more.
The aggregate follows the fleet’s day, ten alarms across a hard urban shift, a fraction of that on a quiet route. Multiplied out, alarm evidence tends to land in the low single gigabytes per vehicle per month, a figure the arithmetic section already walked. The total is modest precisely because the system uploads moments, never hours.
The platform’s alarm policy tunes the parcel per type. A hard-braking event might send stills alone; a collision-grade trigger sends the full clip set; the policies live server-side, pushed to terminals, so a fleet can cheapen the minor types without touching the major ones. The tuning belongs to the safety owner, with the data manager consulted, in that order.
Failed uploads have a data shape of their own. A parcel that dies mid-transfer on a weak cell resumes from where it stopped at the next attempt, the resumable design the attachment flow carries, so the retry costs the missing remainder, never the whole parcel again. A vehicle parked in a coverage hole all evening sends nothing, holds everything, then clears its backlog in the first minutes of good signal, a burst that explains the day-level spikes a usage portal sometimes shows. The burst is honest traffic; the lesson is to read usage by week, never by single day.
This is also the traffic a fleet should fence off from its own cost-cutting. The evidence parcel is what regulators and insurers read after a collision; a configuration that thins it to save megabytes saves trivially and risks the only data that ever justifies its cost. Trim the live habits first, then the playback habits; the evidence is not a place to save money.
Underneath video runs the telemetry the base protocol carries: registration when the link comes up, the heartbeat that keeps the session alive, the position report a few times a minute, the status and alarm flags inside it. Each message is tens of bytes inside a small frame. Registration adds a burst of a few frames at each link re-establishment, dozens of times a day in patchy coverage, still megabytes at month scale.
Summed across a month, this baseline traffic lands in the tens to low hundreds of megabytes, shaped by the reporting cadence the platform sets. Against the video tiers it is a rounding error; the one mistake to avoid is letting an integrator blame a fat bill on telemetry, a claim the arithmetic dismisses in one line: a position report of a few hundred bytes every ten seconds comes to under a hundred megabytes a month, with slower cadences shrinking it further.
Three settings decide more of the bill than every other menu combined, which makes them the shortest worthwhile audit in the configuration. The sub-stream bitrate sets the price of a viewing minute. The resolution tier behind it sets the floor that bitrate can reach before the picture stops being useful. The alarm clip length sets the size of every evidence parcel the fleet will ever send.
Each has a sensible band. A sub stream in the few-hundred-kilobit range carries a usable supervisory picture; pushing it lower buys little money and costs the picture its point, pushing it toward a megabit doubles spend for sharpness a control room rarely needs. Clip length follows the evidence requirement, the seconds around the trigger that reconstruct the event, with anything past that padding the parcel.
These settings are tariff decisions presented as technical menus. They belong in the same review as the data contract, set once with the arithmetic open on the table, then changed deliberately, with the expected effect on the monthly figure written next to the change.
Ownership of the settings decides whether they hold. Stream rates and clip lengths commonly sit in platform-pushed parameter sets, the remote-configuration channel the base protocol provides, which means a carefully chosen profile can be applied fleet-wide in an afternoon, then drift back as depot technicians adjust single vehicles by hand. A quarterly diff of running parameters against the approved profile keeps those choices from drifting.
One more setting hides outside the video menus: the reporting cadence of the telemetry itself. Platforms tighten position reporting for vehicles in sensitive service, five-second cadences or finer, then forget to relax them; the baseline doubles unnoticed. The arithmetic stays small in absolute terms, the principle stays the same, every interval setting somewhere is a recurring charge against the SIM. A parameter survey twice a year, cadence included, keeps those charges deliberate.
Runaway bills repeat the same few stories. The forgotten wall: live tiles opened for an incident, left running for a week. The full-rate habit: playback or live viewing configured at main-stream rates because somebody preferred the sharper picture in January; nobody looked again. The misaimed upload: a terminal set to push recordings on a schedule, a feature meant for depot WiFi, left active on the mobile link. Cross-border work adds the roaming multiplier, the same megabytes at several times the home rate, a line that needs its own alert threshold.
Each story shares a shape: a human choice, reasonable on the day it was made, billing silently from then on. The system does not warn, the carrier counts, the surprise arrives monthly. The defence is plain, a standing cap per SIM at the carrier, an alert threshold under it, a monthly glance at the per-vehicle spread.
The spread is the diagnostic. A fleet’s vehicles doing similar work should bill within sight of each other; one SIM running at multiples of the rest is a configuration error or a misuse with a vehicle number attached. Finding it costs five minutes with the carrier’s portal, which is the cheapest audit in the whole monitoring stack.
Pooled data plans soften overage charges and blunt that diagnostic. A pool absorbs one runaway SIM into the fleet average until the whole pool breaches, weeks later than a per-SIM alert would have fired. Keeping per-SIM visibility inside a pooled contract, a view carriers commonly expose on request, costs nothing beyond asking for the report.
Caps need one caution attached. A SIM that hits a hard ceiling mid-month stops carrying alarms with everything else, so the cap that protects the budget can cut off the evidence channel exactly when a bad month produces the events that need uploading. The safer designs cap with headroom above the worst honest month, alert early, then leave the choke to a human who can see why the spend rose.
Assembled for one composite vehicle, the pieces show what a month looks like. Assume a working profile: telemetry at a ten-second cadence, eight alarms a day at clip-and-stills parcels, forty minutes of sub-stream viewing on an average day, one half-hour remote playback session a week at a lowered rate. No exotic habits, no standing wall.
The arithmetic prices it line by line. Telemetry, under a tenth of a gigabyte. Alarm parcels, around one and a half. Viewing, forty minutes a day at 512 kilobits, about four and a half across the month. The weekly playback sessions add roughly another gigabyte. The composite lands near seven gigabytes, with the discretionary viewing line the largest block, the usual finding.
Two-thirds of the seven is one habit: the daily forty minutes of watching.
The number is a working estimate, never a quote. Real fleets sit above or below it by their habits, which is the point of building the sheet: each line maps to a habit someone controls, with a name attached and a setting behind it. The field-measured totals across live fleets, with their seasonal swings, are a measurement exercise in their own right; this sheet is the tool that makes those measurements explainable when they arrive. Seasonality belongs on the sheet’s margin. Viewing hours climb in incident-heavy months, winter typically; alarm counts follow weather and daylight; playback follows claims. A sheet that carries last winter’s multiplier next to the summer baseline stops the December bill from reading as a fault.
The monthly bill is the system’s own usage record, kept in numbers everyone reads. Map its lines to the three spenders: a baseline of megabytes is the telemetry; low steady gigabytes match the alarm cadence the platform logged; everything above that is viewing and playback, the discretionary tier, attributable to the control room’s habits. Two or three months of that mapping leave a fleet with its own consumption model, calibrated better than any vendor estimate, ready to price the next contract renewal from evidence. That model, plus the one conversion rule, is the entire toolkit this subject requires.
Reconcile the two counters once a quarter. The carrier’s portal counts billable bytes; the platform logs sessions, alarms, viewing minutes. They will never match to the megabyte, overheads see to that; they should match in shape. A carrier line with no platform story behind it is the finding that pays for the whole exercise.
Naming discipline makes the reading fast. SIMs labelled by fleet number in the carrier portal turn the per-vehicle spread into a one-glance table; anonymous ICCID strings turn the same exercise into an afternoon of cross-referencing. A spreadsheet column mapping ICCID to fleet number does the same job where a portal lacks label fields. The five minutes a fitter spends registering the label at installation saves time at every bill. The label outlives staff turnover too: the question of which vehicle a SIM belongs to gets asked again every time the team changes, answered once if the answer lives in the portal.
For a buyer the sequence runs: count the streams the operation genuinely watches, price them by the hour with the conversion rule, add the alarm cadence at parcel size, add a telemetry allowance, then buy the plan one comfortable step above the total with a soft alert at the boundary. A fleet that does this on one sheet of paper before the contract meeting stops buying data as a guess and starts buying it by a known burn rate it controls. The same sheet, kept in the contract folder and revisited at renewal, turns the next negotiation from guesswork into a reading of twelve known months.
At a typical sub-stream rate of 512 kilobits per second, about a quarter of a gigabyte per hour. At a 1080p main-stream rate near 4 megabits, close to 1.8 gigabytes. The conversion is fixed: 1 megabit per second held for an hour moves 0.45 gigabytes, scaling linearly with bitrate. The figure also scales with channels, so four channels watched together cost four times the single-channel rate.
No. The full recording stays on the vehicle’s storage at main-stream quality. What crosses 4G is the sub-stream copy when someone watches live, the short evidence parcel when an alarm fires, plus the small telemetry the base protocol carries. Continuous full upload would cost tens of gigabytes per day per vehicle; the card, sized in the hundreds of gigabytes, does the heavy storage, with the SIM carrying slices measured in megabytes.
Little. A 20-second clip at 2 megabits is around 5 megabytes, with photographs adding fractions more. Even an alarm-heavy urban vehicle tends to spend only low gigabytes a month on evidence, which is why this traffic should be protected from cost-cutting, never trimmed.
The sub-stream bitrate, because it prices every viewing minute. After it, the alarm clip length, which sizes every evidence parcel. Both are tariff decisions and belong in the same review as the data contract itself. Resolution sits behind the bitrate as the floor-setter, third in line.
Vehicles doing similar work should bill similarly. An outlier SIM usually means a configuration error, a live view left open on that vehicle, full-rate playback habits, or a scheduled upload meant for depot WiFi running over 4G. The carrier portal’s per-SIM view finds it in minutes. Pooled plans hide the outlier until the pool itself breaches, so keep the per-SIM view even inside a pool.
Cap with care. A hard ceiling reached mid-month cuts off alarm evidence along with everything else, exactly when a bad month produces the events that need uploading. Safer practice sets the cap above the worst honest month, alerts early at a soft threshold, then leaves the cut-off decision to a person.
