Combined RSSI and ranging position estimators

This documentation was generated with the assistance of AI. Please report any inaccuracies.

com.irurueta.navigation.indoor.position also provides estimators that combine RSSI and ranging readings in a single position estimate. There are two distinct families, differing only in what kind of readings a fingerprint is allowed to hold:

RangingAndRssi: every reading is a single paired RangingAndRssiReading — a measurement that carries both a ranging distance and an RSSI value for the same source at the same instant.
Mixed: a fingerprint may hold an arbitrary, unpaired mixture of RangingReading, RssiReading, and/or RangingAndRssiReading instances — some sources may have only ranging, others only RSSI, others both, in any count. As the library’s own documentation puts it, Mixed estimators are like their RangingAndRssi counterparts "but allow mixing different kinds of readings."

Class hierarchy

classDiagram class PositionEstimator~P~ { <<abstract>> } class RangingAndRssiPositionEstimator~P~ { <<abstract>> } class MixedPositionEstimator~P~ { <<abstract>> } class RobustPositionEstimator~P~ { <<abstract>> } class RobustRangingAndRssiPositionEstimator~P~ { <<abstract>> } class RobustMixedPositionEstimator~P~ { <<abstract>> } class SequentialRobustRangingAndRssiPositionEstimator~P~ class SequentialRobustMixedPositionEstimator~P~ note for SequentialRobustRangingAndRssiPositionEstimator "coarse: RSSI, refine: ranging" note for SequentialRobustMixedPositionEstimator "degrades gracefully if one type is scarce" PositionEstimator <|-- RangingAndRssiPositionEstimator RangingAndRssiPositionEstimator <|-- LinearRangingAndRssiPositionEstimator RangingAndRssiPositionEstimator <|-- NonLinearRangingAndRssiPositionEstimator RobustPositionEstimator <|-- RobustRangingAndRssiPositionEstimator RobustRangingAndRssiPositionEstimator <|-- RANSACRobustRangingAndRssiPositionEstimator2D RobustRangingAndRssiPositionEstimator <|-- LMedSRobustRangingAndRssiPositionEstimator2D RobustRangingAndRssiPositionEstimator <|-- MSACRobustRangingAndRssiPositionEstimator2D RobustRangingAndRssiPositionEstimator <|-- PROSACRobustRangingAndRssiPositionEstimator2D RobustRangingAndRssiPositionEstimator <|-- PROMedSRobustRangingAndRssiPositionEstimator2D PositionEstimator <|-- MixedPositionEstimator MixedPositionEstimator <|-- LinearMixedPositionEstimator MixedPositionEstimator <|-- NonLinearMixedPositionEstimator RobustPositionEstimator <|-- RobustMixedPositionEstimator RobustMixedPositionEstimator <|-- RANSACRobustMixedPositionEstimator2D RobustMixedPositionEstimator <|-- LMedSRobustMixedPositionEstimator2D RobustMixedPositionEstimator <|-- MSACRobustMixedPositionEstimator2D RobustMixedPositionEstimator <|-- PROSACRobustMixedPositionEstimator2D RobustMixedPositionEstimator <|-- PROMedSRobustMixedPositionEstimator2D

Every class above has a 2D and 3D form (LinearRangingAndRssiPositionEstimator2D/3D, RANSACRobustMixedPositionEstimator3D, etc.); SequentialRobustRangingAndRssiPositionEstimator and SequentialRobustMixedPositionEstimator stand outside the RobustPositionEstimator hierarchy entirely — they are self-contained classes that internally compose two other estimators (see Sequential two-stage estimation).

How readings are combined

The two measurement types are unified by PositionEstimatorHelper, shared by every linear, non-linear, and robust estimator in both families. For each reading, depending on its ReadingType:

RANGING_READING

one distance row:

RSSI_READING

one distance row, converted through the path-loss model

RANGING_AND_RSSI_READING

two distance rows for the same source position: one from the ranging distance, one from the RSSI path-loss conversion

There is no special joint equation — an RSSI value is simply converted to a plain Euclidean distance first, and every resulting (position, distance) pair is then stacked as an ordinary trilateration row, indistinguishable from a pure-ranging row once built. This is what "combining" means in both the linear and non-linear estimators: more rows in the same system, not a different system.

Linear estimation

LinearRangingAndRssiPositionEstimator and LinearMixedPositionEstimator both call PositionEstimatorHelper.buildPositionsAndDistances(…) to pool every reading into (position, distance) rows as described above, then hand the pooled arrays to the same homogeneous/inhomogeneous linear trilateration solvers used by the RSSI and ranging hierarchies (useHomogeneousLinearSolver, default true). The linear system itself is agnostic to which physical phenomenon produced each row.

Non-linear estimation

NonLinearRangingAndRssiPositionEstimator and NonLinearMixedPositionEstimator call PositionEstimatorHelper.buildPositionsDistancesAndDistanceStandardDeviations(…), again emitting two triples per paired reading, with variance propagated per row (combining transmit-power, received-power, path-loss exponent, and optional source-position covariance for RSSI-derived rows). Levenberg-Marquardt then minimizes one joint weighted cost over every row regardless of origin:

RSSI-derived rows typically carry a larger (reflecting extra path-loss-model uncertainty) than ranging-derived rows, so they are automatically down-weighted relative to direct ranging measurements — with no special-cased term needed in the cost function itself.

Robust estimation

Plain robust estimators

RobustRangingAndRssiPositionEstimator and RobustMixedPositionEstimator run a single robust lateration solver over the entire pooled row list — ranging-derived and RSSI-derived rows mixed together in one RANSAC/LMedS/MSAC/PROSAC/PROMedS sampling loop, exactly as described for the RSSI and ranging hierarchies. Thresholds and quality scores are not separated by measurement type — by the time data reaches the robust solver it is one homogeneous list of (position, distance, [, qualityScore]) rows with no memory of which physical phenomenon produced each one.

The one place reading type still matters is when evenlyDistributeReadings (default true) re-ranks sampling priority via ReadingSorter, which breaks quality-score ties in a fixed order:

i.e. pure ranging rows are preferred first, paired rows next, pure RSSI rows last — affecting sampling order only, not the single scalar inlier threshold shared by the whole dataset.

Sequential two-stage estimation

SequentialRobustRangingAndRssiPositionEstimator and SequentialRobustMixedPositionEstimator are not RobustPositionEstimator subclasses — they are self-contained classes that internally compose two independent robust estimators in sequence: RSSI first for a coarse estimate, then ranging to refine it.

flowchart LR A[Paired/mixed fingerprint] --> B[Split into pure RSSI<br/>and pure ranging fingerprints] B --> C["Stage 1: RobustRssiPositionEstimator<br/>(own robust method, threshold, quality scores)"] C --> D[Coarse position estimate] D --> E["Stage 2: RobustRangingPositionEstimator<br/>seeded with coarse position as initial estimate"] E --> F[Final refined position + covariance]

For the paired RangingAndRssi family, the fingerprint is split deterministically (every reading yields both a ranging and an RSSI sample), so both stages always run. For the Mixed family, each stage only runs if enough readings of that type are present (numRangingReadings/numRssiReadings each meet the minimum required sources); Mixed degrades gracefully:

If ranging readings are insufficient, the RSSI-only coarse result is returned directly as the final answer.
If RSSI readings are insufficient, initialPosition (or the centroid) seeds the ranging-only stage directly.
isReady() only requires one of the two stages to be available.

Every configuration knob is duplicated per stage with a ranging/rssi prefix — separate robust method (both default PROMedS), confidence, max iterations, threshold, linear-solver flags, and quality scores — so each stage can be tuned independently even though both eventually contribute to one position.

Usage parameters

sources, fingerprint (RangingAndRssiFingerprint for the paired family, generic Fingerprint for Mixed), listener — common to all variants.
Linear: useHomogeneousLinearSolver (default true).
Non-linear: initialPosition (else centroid), useRadioSourcePositionCovariance, fallbackDistanceStandardDeviation, getCovariance().
Robust (plain, common): confidence (default 0.99), maxIterations (default 5000), preliminarySubsetSize, refineResult/keepCovariance (default true), evenlyDistributeReadings (default true), plus per-algorithm threshold/stopThreshold and, for PROSAC/PROMedS, sourceQualityScores / fingerprintReadingsQualityScores.
Sequential (per stage, ranging/rssi prefixed): rangingRobustMethod/rssiRobustMethod (default PROMedS each), rangingConfidence/rssiConfidence, rangingMaxIterations/rssiMaxIterations, rangingThreshold/rssiThreshold, useRangingLinearSolver/useRssiLinearSolver, evenlyDistributeRangingReadings/evenlyDistributeRssiReadings; shared initialPosition, refineResult, keepCovariance, progressDelta.

Static create(…) factories on RobustRangingAndRssiPositionEstimator2D/3D and RobustMixedPositionEstimator2D/3D build any plain robust variant from a RobustEstimatorMethod enum.

API reference

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

Class Source Javadoc

Class	Source	Javadoc
`Fingerprint`	Source	Javadoc
`LMedSRobustMixedPositionEstimator2D`	Source	Javadoc
`LMedSRobustRangingAndRssiPositionEstimator2D`	Source	Javadoc
`LinearMixedPositionEstimator`	Source	Javadoc
`LinearRangingAndRssiPositionEstimator`	Source	Javadoc
`LinearRangingAndRssiPositionEstimator2D`	Source	Javadoc
`LinearRangingAndRssiPositionEstimator3D`	Source	Javadoc
`MSACRobustMixedPositionEstimator2D`	Source	Javadoc
`MSACRobustRangingAndRssiPositionEstimator2D`	Source	Javadoc
`MixedPositionEstimator`	Source	Javadoc
`NonLinearMixedPositionEstimator`	Source	Javadoc
`NonLinearRangingAndRssiPositionEstimator`	Source	Javadoc
`PROMedSRobustMixedPositionEstimator2D`	Source	Javadoc
`PROMedSRobustRangingAndRssiPositionEstimator2D`	Source	Javadoc
`PROSACRobustMixedPositionEstimator2D`	Source	Javadoc
`PROSACRobustRangingAndRssiPositionEstimator2D`	Source	Javadoc
`PositionEstimator`	Source	Javadoc
`PositionEstimatorHelper`	Source	Javadoc
`RANSACRobustMixedPositionEstimator2D`	Source	Javadoc
`RANSACRobustMixedPositionEstimator3D`	Source	Javadoc
`RANSACRobustRangingAndRssiPositionEstimator2D`	Source	Javadoc
`RangingAndRssiFingerprint`	Source	Javadoc
`RangingAndRssiPositionEstimator`	Source	Javadoc
`RangingAndRssiReading`	Source	Javadoc
`RangingReading`	Source	Javadoc
`ReadingSorter`	Source	Javadoc
`ReadingType`	Source	Javadoc
`RobustMixedPositionEstimator`	Source	Javadoc
`RobustMixedPositionEstimator2D`	Source	Javadoc
`RobustMixedPositionEstimator3D`	Source	Javadoc
`RobustPositionEstimator`	Source	Javadoc
`RobustRangingAndRssiPositionEstimator`	Source	Javadoc
`RobustRangingPositionEstimator`	Source	Javadoc
`RobustRssiPositionEstimator`	Source	Javadoc
`RssiReading`	Source	Javadoc
`SequentialRobustMixedPositionEstimator`	Source	Javadoc
`SequentialRobustRangingAndRssiPositionEstimator`	Source	Javadoc

Fingerprint