KPIs That Matter for RTK: Key Metrics for High-Performance GNSS

Estimated reading time: 9 minutes

Summary

RTK performance is proven with measurable KPIs: accuracy, availability, integrity, continuity, time-to-fix, fix rate, latency/age of corrections, and satellite geometry. This guide does two things most KPI lists skip: it names the field that carries each number, and who published the threshold you measure against. That is the difference between a metric you can quote and one you can only assert.

Key takeaways

• Accuracy (e.g., RMS/R95) and availability show whether your solution meets spec across the day, not just at a single point in time.
• Integrity and continuity protect safety-critical work by detecting bad data and avoiding unexpected dropouts.
• Time-to-fix under 10 seconds is the published multi-constellation figure; u-blox lists 30 seconds for GPS alone, so a habitual 30 second wait is a diagnosis, not a target.
• Correction age near 1 second is the target, 2 seconds the ceiling NGS recommends. Integer fix rate has no published pass mark; set it from your own baseline.
• Use dashboards and logs to monitor KPIs continuously; verify against ground truth or control points.

What Makes a Number a KPI

GNSS provides global positioning using multiple satellite constellations. RTK adds real-time corrections from a base or network to achieve centimeter-level accuracy. The same handful of numbers decides whether a crew can trust a stake, a flight line or a guidance pass, and whether a disagreement with a provider is settled with evidence or opinion.

A number earns the label KPI when it passes three tests: you can read it off equipment you already own, it is comparable day to day, and it can be checked against a published threshold rather than a felt one. Most RTK metrics fail the third test.

Rover KPIs: Accuracy, Availability, Integrity, Continuity

These four are the standard parameters for specifying any positioning service, and the four a correction stream has to be judged on. Everything else below is evidence you collect to prove them. For the mechanism underneath, see how ambiguity resolution produces a centimeter result. Track them over time:

• Accuracy: horizontal/vertical error (RMS, CEP, R95). Validate with control points or independent ground truth.
• Availability: % of time the solution meets accuracy spec in the work window. This is the KPI most teams skip and the one that decides whether a day was billable.
• Integrity: ability to detect and exclude bad measurements; define alert limits and protection levels for your use case. Borrowed from aviation practice, so treat it as a hard requirement only where a wrong position is a safety event.
• Continuity: resistance to dropouts across a task (e.g., uninterrupted operation over N minutes).
• Time-to-Fix: median time from start/restart to RTK Fix. Report the median and the 95th percentile, never the best run.
• Fix Rate: share of epochs with integer ambiguity resolution (Fix vs Float/Single).
• Latency & Age of Corrections: end-to-end stream latency and rover-reported correction age.
• Geometry & Signal: PDOP/HDOP, satellites in view/used, SNR, cycle-slip counts, multipath indicators.

Where to Read Each KPI in the NMEA Stream

Every KPI above is already in the sentences your receiver emits. Logging them needs a serial capture, not a vendor dashboard. Field names follow the u-blox F9 HPG interface description (retrieved September 2026); other vendors emit the same NMEA sentences, so the names carry over.

KPI	Sentence	Field	What it contains
Fix state, and fix rate	GGA	quality	0 no fix, 1 autonomous, 2 differential, 4 RTK fixed, 5 RTK float
Correction age	GGA	diffAge	Age of differential corrections in seconds, null when none are in use
Satellites used	GGA	numSV	Count of satellites in the position solution
Horizontal geometry	GGA	HDOP	Horizontal dilution of precision
Full geometry	GSA	PDOP, VDOP	Position and vertical dilution of precision
Position error estimate	GST	stdLat, stdLong, stdAlt	Standard deviations of latitude, longitude and altitude error, in meters
Signal strength	GSV	cno	Carrier to noise density per satellite, in dBHz

Two notes. Fix rate is not a field but a calculation: the share of logged epochs whose GGA quality equals 4. And field indexes differ between vendor documents, since u-blox counts the message identifier as field 0, so match on the name, not the number.

What Counts as Good, and Who Published the Number

Time-to-fix. u-blox publishes RTK convergence for the ZED-F9P at under 10 seconds for every multi-constellation mode it lists, and under 30 seconds for GPS alone (ZED-F9P-05B data sheet, Table 2, retrieved September 2026). Emlid quotes about 5 seconds typical for the Reach RS3. So 30 seconds is not a target; it is what the same silicon does on one constellation. u-blox notes it depends on atmosphere, baseline length, antenna, multipath and geometry, which is why you track a distribution. Our own documentation asks users in marginal conditions to allow 60 to 90 seconds before assuming a fault: 10 seconds is the specification, 90 seconds is patience under a tree.

Correction age. The NGS single-base guidelines recommend using data with latencies no greater than 2 seconds (retrieved September 2026). RTKdata publishes the same bands: full performance under 1 second, degraded with more float between 1 and 3 seconds, stale beyond 3. Target 1 second, alarm at 2, escalate at 3.

Fix rate. There is no published pass mark. Fix rate is dominated by sky view, so set yours from a baseline week on your own sites and treat departures from it as the signal. The number often mistaken for a threshold is a different KPI: NGS records that manufacturers state 99.9 percent confidence in a correct on-the-fly integer fix. That is fix reliability, whether a fix is right, not fix rate, how often you have one.

Accuracy. NGS publishes acceptance criteria per class. They date from 2011, target single-base work, and NGS notes newer hardware may reach the tighter class on looser criteria. Read them as a floor.

Class	Typical precision (2 sigma)	Baseline (max)	Observation	PDOP (max)	Satellites (min)	RMS (max)
RT1	0.01-0.02 m H, 0.02-0.04 m V	10 km	180 epochs at 1 s	2.0	7	0.01 m
RT2	0.02-0.04 m H, 0.03-0.05 m V	15 km	12 epochs at 5 s	3.0	6	0.015 m
RT3	0.04-0.06 m H, 0.04-0.08 m V	20 km	15 epochs at 1 s	4.0	5	0.03 m
RT4	0.1-0.2 m H, 0.1-0.3 m V	any, if fixed	10 epochs at 1 s	6.0	5	0.05 m

Accuracy also falls off with distance from the reference station, a separate subject covered in how baseline distance changes the result.

Provider KPIs: What to Require From a Correction Network

Choosing a network is a different job from monitoring the one you pay for; selection criteria are covered in how network corrections are generated and compared. These are the indicators worth watching once you have signed:

• Reference network health: station uptime, spacing, monitoring, and quality alarms.
• Stream uptime: mountpoint availability, reconnect success, authentication errors.
• Message coverage: required RTCM types present and timely (observations, MSM, antenna/geo, bias/orbit models).
• Latency distribution: p50/p95 latency for your region and access method (cellular, Wi-Fi, radio).
• Regional performance: KPI heatmaps by route/site and time-of-day.

Latency is the one with a published budget on the provider's side of the wire. The NGS network guidelines expect under 1 second from reference station to central processing, and call 2.5 seconds a highly undesirable limit (retrieved September 2026), before a single byte reaches your rover. Useful in a service conversation because it is neither yours nor ours.

Diagnostic KPIs: Constellations, DOP and Signal Strength

When an outcome KPI goes bad, these four say why:

• Constellation/band mix: GPS/Galileo/GLONASS/BeiDou across L1/L2/L5; more diversity improves resilience. On the same ZED-F9P, adding one constellation to GPS moves published convergence from under 30 seconds to under 10.
• PDOP/HDOP & SNR: track trends alongside environment changes (trees, buildings, bridges). NGS gives 30 dB as a common floor for usable signal to noise.
• Cycle slips & multipath: indicators for antenna placement or RF interference issues. The failure mode itself is treated in depth in why a fix collapses in urban canyons.
• Correction age & Fix/Float transitions: correlate with network coverage and motion events.

How RTKdata Reports These Numbers

RTKdata operates 20,000+ reference stations across 140+ countries. Inside dense coverage the working figure is 1-2 cm horizontal and 2-3 cm vertical. The coverage map splits that into two published zones, so an area can be checked before a crew is sent: a 2 cm zone where RTK FIX is reliable at survey grade, and a 2-10 cm zone where FIX is achievable and convergence takes longer. The correction-age bands quoted earlier are ours too, published, and the same standard this article applies to everyone else.