GNSS receivers — from metres to centimetres

All GNSS receivers solve the same equations; they differ in how many frequencies and constellations they track, what corrections they can apply, how gracefully they degrade under interference, and what they give you for tight coupling with an IMU.

Accuracy modes

Mode	Accuracy	Needs	Convergence
Single-frequency SPS	3–5 m	nothing	instant
SBAS (WAAS/EGNOS)	1–2 m	geostationary correction signal	instant
Dual-frequency	1–2 m	L1+L5 capable receiver	instant (removes most ionosphere)
RTK	1–2 cm	base station / network within ~20 km, data link	seconds once "fixed"
PPP	2–10 cm	precise orbit/clock stream (often subscription)	minutes to converge

Dual-frequency matters even without corrections because the ionospheric delay is dispersive: measuring at two frequencies lets you form a combination that cancels most of it. It is usually the single best accuracy-per-dollar upgrade.

What to look for beyond accuracy

Carrier-phase output and raw observables — mandatory if you want to do your own RTK/PPP or tightly-coupled GNSS/INS rather than consuming a filtered position.
PPS output and time-tag quality. A hardware pulse-per-second is what lets you align GNSS with IMU samples to microseconds; without it, latency becomes an unmodelled bias.
Update rate. 1 Hz is fine for a filter's aiding; 10–20 Hz helps dynamic platforms.
Multi-constellation (GPS + Galileo + GLONASS + BeiDou) — mostly about availability and geometry in obstructed environments, i.e. improving GDOP.
Interference resilience: spectrum monitoring, adaptive notch filters, and reporting of jamming and spoofing indicators. For a real threat model use a CRPA (controlled-reception-pattern antenna).
The antenna is half the system. A survey-grade antenna with a good ground plane and stable phase centre matters as much as the receiver; for RTK you must know the phase-centre offset.

Do not treat the receiver's position output as truth. It is already the product of an internal filter with its own assumptions. For tight coupling, consume pseudoranges and carrier phase so your filter can weight and gate each satellite — and reject spoofs — itself.

Why the extra satellite exists

The receiver clock bias is common to every measurement, which is precisely what makes it separable — and why you need one more satellite than you have position unknowns. Turn "solve for δt" off below and watch the circles stop agreeing:

And geometry, not signal strength, is what usually limits the fix:

Representative receivers

Receiver	Class	Notes
u-blox NEO-M9N	consumer, single-freq	cheap, ubiquitous, 2–3 m class
u-blox ZED-F9P	dual-freq RTK	cm-level at low cost; the hobby/robotics standard
Septentrio mosaic-X5	multi-band, anti-jam	strong interference monitoring & mitigation
NovAtel OEM7 series	professional GNSS/INS	integrated INS options, well-documented raw output
Trimble BD9xx	survey grade	reference-quality, long track record
Hemisphere / Tersus boards	mid-range RTK	good price/performance for UAV mapping