Positioning · Navigation · Timing
Understand how the world knows where it is.
From first principles to cutting-edge research — through interactive learning, engineering tools, and real-world applications. Built to cultivate understanding, not just deliver information.
What would you like to do today?
Concepts
Signal Acquisition & Tracking
Every measurement in this course — pseudorange, carrier phase, Doppler — assumed the receiver had already locked onto the satellite. This is how that actually happens. The signal arrives below the noise floor and Doppler-shifted by kilometres per second of satellite motion, so the receiver runs a two-dimensional search over code phase and frequency until a correlation peak erupts from the noise, then hands the peak to tracking loops that follow it continuously, regenerating the code and carrier measurements every millisecond — and, in the process, giving carrier phase its integer ambiguity.
intermediatecoreCarrier Phase & Ambiguity
Under the code rides the raw carrier wave — a ruler with 19-centimetre teeth that a receiver can read to a millimetre. The catch is maddening: you can measure the fraction of a wavelength perfectly, but you have no idea how many whole wavelengths you're standing on. Resolve that integer, and you unlock centimetre positioning.
advancedcoreCoordinate Frames
Geodetic, ECEF, and ENU are three different languages for the exact same point. Knowing which one you're speaking — and how to translate — is the quiet backbone of all GNSS.
beginnercoreDilution of Precision
The same measurements can give you a crisp fix or a mushy one depending purely on the geometry of the satellites overhead. DOP is the number that captures that difference.
beginnercoreThe GNSS Error Budget
Every positioning lesson has leaned on the same short list of villains — satellite clock and orbit, ionosphere, troposphere, multipath, receiver noise — without ever sizing them. This is the accounting. Each source is a range error in metres; corrections shrink some to nothing and leave others untouchable. Add the survivors in quadrature and you get the UERE, which the geometry then multiplies into your position error.
intermediatecoreThe EKF & GNSS/INS Integration
The linear Kalman filter needed a ready-made position fix and a straight-line sky. Real GNSS gives neither: pseudoranges are a nonlinear function of position, and the sky disappears in tunnels and urban canyons. The Extended Kalman Filter linearises each measurement about the current estimate, letting it consume raw ranges directly; and by driving its prediction with an inertial unit, it coasts through GNSS outages — the two sensors covering each other's gaps.
advancedcoreInteger Ambiguity Resolution & LAMBDA
Carrier phase promises millimetre ranging, but only once you know the integer number of whole wavelengths — the ambiguity. RTK and PPP hand you a real-valued estimate of those integers plus their covariance; rounding each one independently fails because they are strongly correlated. LAMBDA reframes fixing as integer least squares, decorrelates the problem with a volume-preserving integer change of basis so the search is fast, finds the best lattice point, and validates it with a ratio test — the step that turns a float solution into a centimetre fix, or wisely declines to.
advancedcoreIntegrity & RAIM
Every lesson so far chased accuracy: getting the fix close to the truth. Safety-of-life GNSS needs something harder — integrity: the fix must come with a trustworthy bound on its own error and a timely alarm when that bound could be exceeded. RAIM delivers both from redundancy alone: extra satellites make the measurements check each other, residuals expose a faulty one, and the geometry turns an undetectable bias into a protection level you compare to an alert limit.
advancedcoreThe Ionosphere & Multi-Frequency
The ionosphere is the biggest error in GNSS — metres of range, and worse near the horizon. The error budget listed it as a line item and PPP used an "iono-free combination" we never justified. Both rest on one beautiful fact: the ionosphere is dispersive, delaying the signal by an amount proportional to 1/f². A single-frequency receiver must model the delay and keep half of it; a dual-frequency receiver measures it directly from the gap between two frequencies and cancels it almost completely — paying only a fixed noise amplification. This is the physics behind that 2.9.
intermediatecoreRecursive Estimation & the Kalman Filter
Every solver so far answers "where am I right now?" from a single snapshot, throwing away everything it knew a second ago. A moving receiver can do far better by remembering. The Kalman filter carries a running belief forward in time, predicts with a motion model, and optimally blends each noisy fix against that prediction by their relative confidence — smoothing the track, estimating velocity it never measured, and reporting its own uncertainty.
advancedcoreMulti-Constellation GNSS
Modern receivers track GPS, Galileo, BeiDou, and GLONASS together — but you cannot just pour them all into one trilateration. Each constellation keeps its own system time and its own reference frame, so its pseudoranges carry a different clock offset. Mix them blindly and the fix goes wrong; the right move is to estimate one receiver-clock unknown per system, absorbing the inter-system bias. That costs one extra satellite per constellation, but buys a sky full of geometry — the difference between a dead receiver and a working one in an urban canyon.
intermediatecorePrecise Point Positioning (PPP)
RTK buys centimetres by borrowing a nearby friend's errors. PPP throws the friend away. With a single receiver anywhere on Earth, it applies global precise orbit and clock products and then models every remaining error explicitly — ionosphere, troposphere, tides, antenna effects — solving for position, clock, and float ambiguities in one estimator. The price is convergence time; the payoff is global decimetre-to-centimetre accuracy with no base at all.
advancedcorePseudorange
Why the ranges in satellite positioning are politely called "pseudo" — and how one humble, shared clock error turns clean geometry into a four-unknown puzzle.
beginnercoreReal-Time Kinematic (RTK)
Your phone is good to a few metres; a survey rover is good to a couple of centimetres — from the same satellites and the same sky. The difference is a second receiver sitting still on a known point. Double-difference their measurements and the shared errors vanish, leaving an integer-ambiguity problem whose solution is a centimetre-level baseline.
advancedcoreSatellite Orbits
A GNSS satellite isn't hovering — it's falling, endlessly, and missing the Earth because it's moving sideways fast enough. Kepler's three laws turn that idea into the numbers that decide altitude, period, speed, and where a satellite will be when you need it.
intermediatecoreThe GNSS Signal & Code Ranging
A GPS signal arrives weaker than the noise around it, yet your phone finds it in milliseconds. The trick is the pseudorandom code: slide a local replica until it snaps into alignment, and the amount you slid is the travel time. This is where pseudorange measurements are actually born.
intermediatecoreSingle Point Positioning
The capstone of the foundations: how a receiver takes several imperfect pseudoranges and, by least squares, produces a single best position, a clock estimate, and an honest measure of how much to trust the answer.
intermediatecoreGNSS Time Systems
Every position is a timestamp wearing a costume. Meet the timescales GNSS runs on — TAI, UTC, GPST, and their cousins — why the satellites refuse to take leap seconds, and how one mishandled second becomes a 300,000-kilometre mistake.
beginnercoreTrilateration
Finding where you are by measuring how far you are from several known points — the geometric heart of satellite positioning.
beginnercore