Radio source estimators

RSSI-based position estimation assumes the access point’s position and transmitted power are known, and solves for the device’s unknown position. Radio source estimation runs the same physics in reverse: now it’s the receivers' positions that are known — several of them, placed around the area — and it’s the source’s position (plus its transmitted power, and optionally its path-loss exponent) that’s unknown.

Since the source radiates its signal equally in all directions, and that signal weakens with distance in the same predictable way, a receiver that reports a strong signal must be relatively close to the source, and one that reports a weak signal must be relatively far. Combining several such readings — each pulling the estimate toward or away from a receiver depending on how strong the signal was there — is enough to solve for where the source must be for all of them to make sense at once, along with how powerful it must be transmitting to produce exactly those readings.

The same idea works with ranging (distance) readings instead of, or alongside, RSSI: see ranging-based estimation and combining ranging and RSSI below.

Every non-robust and robust class has 2D and 3D forms; robust classes have RANSAC/LMedS/MSAC/PROSAC/PROMedS concrete variants (5 per measurement type per dimension). There is no non-sequential robust Mixed estimator — robustly estimating from a heterogeneous, unpaired reading set is only available through SequentialRobustMixedRadioSourceEstimator. The Sequential* classes stand outside the RobustRadioSourceEstimator hierarchy; they compose two other robust estimators internally (see Sequential two-stage estimation).

RssiRadioSourceEstimator fits the log-distance path-loss model directly via Levenberg-Marquardt — there is no closed-form solution here, unlike position estimation, because the source position, its equivalent transmitted power (antenna gains folded in, since they’re unknown for third-party sources), and the path-loss exponent are all potentially unknown at once:

Three independent flags control what’s actually solved for: positionEstimationEnabled (default true), transmittedPowerEstimationEnabled (default true), pathLossEstimationEnabled (default false — is fixed at 2.0 unless explicitly enabled). The class documentation warns that enabling all three at once "usually achieves inaccurate results" — at most two of the three should be estimated simultaneously, with an initial value supplied for the third. Seven separate fitter/Jacobian setups exist internally, one per enabled/fixed combination.

Defaults for initial values: initialTransmittedPowerdBm falls back to the average RSSI across all readings if unset; initialPosition falls back to the centroid of all reading positions. The minimum number of readings is — one more than the number of unknowns.

RangingRadioSourceEstimator is simpler: since ranging gives distance directly, there’s no path-loss model to invert, and only position is ever estimated (no power, no path-loss exponent) — this is the same trilateration problem as ranging position estimation, just with the roles of "known" and "unknown" swapped. It runs a linear lateration solve first (homogeneous or inhomogeneous, useHomogeneousLinearSolver default true) and then, by default (nonLinearSolverEnabled = true), refines the result with the same Levenberg-Marquardt lateration solver used elsewhere in this library. The minimum reading count is — the classic trilateration minimum.

All three measurement types share the same RANSAC-family scoring machinery, implemented once in RobustRadioSourceEstimator and its concrete subclasses. The residual differs only by measurement type:

If refineResult (default true), the winning candidate is always re-solved with the non-robust inner estimator restricted to its inlier subset, and keepCovariance (default true) retains the resulting covariance. The library-wide default robust method is PROMedS, not RANSAC.

=== Sequential two-stage estimation

SequentialRobustRangingAndRssiRadioSourceEstimator and SequentialRobustMixedRadioSourceEstimator stand outside the RobustRadioSourceEstimator hierarchy and instead run two independently configured robust estimators in sequence — mirroring the two-stage approach used by the position-side Sequential estimators, but here both stages target the same radio source’s unknowns rather than a device position:

The class documentation notes this two-stage approach "might produce more stable positions…​ than [the joint, single-pass] RobustRangingAndRssiRadioSourceEstimator," since ranging-only lateration is a better-conditioned problem to robustify on its own before it anchors the RSSI fit. Quality scores supplied to the sequential estimator are split and forwarded to whichever inner stage uses PROSAC/PROMedS.

Static create(…​) factories exist on the 2D/3D robust classes for each measurement type, building any variant from a RobustEstimatorMethod enum (default PROMedS).

Main classes used on this page, linked to their source code and Javadoc: