AbstractLambdaMethod.java
/* Copyright 2002-2020 CS Group
* Licensed to CS Group (CS) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* CS licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.orekit.estimation.measurements.gnss;
import java.util.SortedSet;
import java.util.TreeSet;
import org.hipparchus.linear.RealMatrix;
import org.hipparchus.util.FastMath;
/** Base class for decorrelation/reduction engine for LAMBDA type methods.
* <p>
* This class is based on both the 1996 paper <a href="https://www.researchgate.net/publication/2790708_The_LAMBDA_method_for_integer_ambiguity_estimation_implementation_aspects">
* The LAMBDA method for integer ambiguity estimation: implementation aspects</a> by
* Paul de Jonge and Christian Tiberius and on the 2005 paper <a
* href="https://www.researchgate.net/publication/225518977_MLAMBDA_a_modified_LAMBDA_method_for_integer_least-squares_estimation">
* A modified LAMBDA method for integer least-squares estimation</a> by X.-W Chang, X. Yang
* and T. Zhou, Journal of Geodesy 79(9):552-565, DOI: 10.1007/s00190-005-0004-x
* </p>
* @author Luc Maisonobe
* @since 10.0
*/
public abstract class AbstractLambdaMethod implements IntegerLeastSquareSolver {
/** Number of ambiguities. */
private int n;
/** Decorrelated ambiguities. */
private double[] decorrelated;
/** Lower triangular matrix with unit diagonal, in row order (unit diagonal not stored). */
private double[] low;
/** Diagonal matrix. */
private double[] diag;
/** Z⁻¹ transformation matrix, in row order. */
private int[] zInverseTransformation;
/** Maximum number of solutions seeked. */
private int maxSolutions;
/** Placeholder for solutions found. */
private SortedSet<IntegerLeastSquareSolution> solutions;
/** {@inheritDoc} */
@Override
public IntegerLeastSquareSolution[] solveILS(final int nbSol, final double[] floatAmbiguities,
final int[] indirection, final RealMatrix covariance) {
// initialize the ILS problem search
initializeProblem(floatAmbiguities, indirection, covariance, nbSol);
// perform initial Lᵀ.D.L = Q decomposition of covariance
ltdlDecomposition();
// perform decorrelation/reduction of covariances
reduction();
// transform the Lᵀ.D.L = Q decomposition of covariance into
// the L⁻¹.D⁻¹.L⁻ᵀ = Q⁻¹ decomposition of the inverse of covariance.
inverseDecomposition();
// perform discrete search of Integer Least Square problem
discreteSearch();
return recoverAmbiguities();
}
/** Initialize ILS problem.
* @param floatAmbiguities float estimates of ambiguities
* @param indirection indirection array to extract ambiguity covariances from global covariance matrix
* @param globalCovariance global covariance matrix (includes ambiguities among other parameters)
* @param nbSol number of solutions to search for
*/
private void initializeProblem(final double[] floatAmbiguities, final int[] indirection,
final RealMatrix globalCovariance, final int nbSol) {
this.n = floatAmbiguities.length;
this.decorrelated = floatAmbiguities.clone();
this.low = new double[(n * (n - 1)) / 2];
this.diag = new double[n];
this.zInverseTransformation = new int[n * n];
this.maxSolutions = nbSol;
this.solutions = new TreeSet<>();
// initialize decomposition matrices
for (int i = 0; i < n; ++i) {
for (int j = 0; j < i; ++j) {
low[lIndex(i, j)] = globalCovariance.getEntry(indirection[i], indirection[j]);
}
diag[i] = globalCovariance.getEntry(indirection[i], indirection[i]);
zInverseTransformation[zIndex(i, i)] = 1;
}
}
/** Get a reference to the diagonal matrix of the decomposition.
* <p>
* BEWARE: the returned value is a reference to an internal array,
* it is <em>only</em> intended for subclasses use (hence the
* method is protected and not public).
* </p>
* @return reference to the diagonal matrix of the decomposition
*/
protected double[] getDiagReference() {
return diag;
}
/** Get a reference to the lower triangular matrix of the decomposition.
* <p>
* BEWARE: the returned value is a reference to an internal array,
* it is <em>only</em> intended for subclasses use (hence the
* method is protected and not public).
* </p>
* @return reference to the lower triangular matrix of the decomposition
*/
protected double[] getLowReference() {
return low;
}
/** Get the reference decorrelated ambiguities.
* @return reference to the decorrelated ambiguities.
**/
protected double[] getDecorrelatedReference() {
return decorrelated;
}
/** Get the maximum number of solutions seeked.
* @return the maximum number of solutions seeked
*/
protected int getMaxSolution() {
return maxSolutions;
}
/** Add a new solution.
* @param fixed solution array
* @param squaredNorm squared distance to the corresponding float solution
*/
protected void addSolution(final long[] fixed, final double squaredNorm) {
solutions.add(new IntegerLeastSquareSolution(fixed, squaredNorm));
}
/** Remove spurious solution.
*/
protected void removeSolution() {
solutions.remove(solutions.last());
}
/** Get the number of solutions found.
* @return the number of solutions found
*/
protected int getSolutionsSize() {
return solutions.size();
}
/** Perform Lᵀ.D.L = Q decomposition of the covariance matrix.
*/
protected abstract void ltdlDecomposition();
/** Perform LAMBDA reduction.
*/
protected abstract void reduction();
/** Find the best solutions to the Integer Least Square problem.
*/
protected abstract void discreteSearch();
/** Inverse the decomposition.
* <p>
* This method transforms the Lᵀ.D.L = Q decomposition of covariance into
* the L⁻¹.D⁻¹.L⁻ᵀ = Q⁻¹ decomposition of the inverse of covariance.
* </p>
*/
protected abstract void inverseDecomposition();
/** Perform one integer Gauss transformation.
* <p>
* This method corresponds to algorithm 2.1 in X.-W Chang, X. Yang and T. Zhou paper.
* </p>
* @param row row index (counting from 0)
* @param col column index (counting from 0)
*/
protected void integerGaussTransformation(final int row, final int col) {
final int mu = (int) FastMath.rint(low[lIndex(row, col)]);
if (mu != 0) {
// update low triangular matrix (post-multiplying L by Zᵢⱼ = I - μ eᵢ eⱼᵀ)
low[lIndex(row, col)] -= mu;
for (int i = row + 1; i < n; ++i) {
low[lIndex(i, col)] -= mu * low[lIndex(i, row)];
}
// update Z⁻¹ transformation matrix
for (int i = 0; i < n; ++i) {
// pre-multiplying Z⁻¹ by Zᵢⱼ⁻¹ = I + μ eᵢ eⱼᵀ
zInverseTransformation[zIndex(row, i)] += mu * zInverseTransformation[zIndex(col, i)];
}
// update decorrelated ambiguities estimates (pre-multiplying a by Zᵢⱼᵀ = I - μ eⱼ eᵢᵀ)
decorrelated[col] -= mu * decorrelated[row];
}
}
/** Perform one symmetric permutation involving rows/columns {@code k0} and {@code k0+1}.
* <p>
* This method corresponds to algorithm 2.2 in X.-W Chang, X. Yang and T. Zhou paper.
* </p>
* @param k0 diagonal index (counting from 0)
* @param delta new value for diag[k0+1]
*/
protected void permutation(final int k0, final double delta) {
final int k1 = k0 + 1;
final int k2 = k0 + 2;
final int indexk1k0 = lIndex(k1, k0);
final double lk1k0 = low[indexk1k0];
final double eta = diag[k0] / delta;
final double lambda = diag[k1] * lk1k0 / delta;
// update diagonal
diag[k0] = eta * diag[k1];
diag[k1] = delta;
// update low triangular matrix
for (int j = 0; j < k0; ++j) {
final int indexk0j = lIndex(k0, j);
final int indexk1j = lIndex(k1, j);
final double lk0j = low[indexk0j];
final double lk1j = low[indexk1j];
low[indexk0j] = lk1j - lk1k0 * lk0j;
low[indexk1j] = lambda * lk1j + eta * lk0j;
}
low[indexk1k0] = lambda;
for (int i = k2; i < n; ++i) {
final int indexik0 = lIndex(i, k0);
final int indexik1 = indexik0 + 1;
final double tmp = low[indexik0];
low[indexik0] = low[indexik1];
low[indexik1] = tmp;
}
// update Z⁻¹ transformation matrix
for (int i = 0; i < n; ++i) {
final int indexk0i = zIndex(k0, i);
final int indexk1i = indexk0i + n;
final int tmp2 = zInverseTransformation[indexk0i];
zInverseTransformation[indexk0i] = zInverseTransformation[indexk1i];
zInverseTransformation[indexk1i] = tmp2;
}
// update decorrelated ambiguities
final double tmp = decorrelated[k0];
decorrelated[k0] = decorrelated[k1];
decorrelated[k1] = tmp;
}
/** Recover ambiguities prior to the Z-transformation.
* @return recovered ambiguities
*/
private IntegerLeastSquareSolution[] recoverAmbiguities() {
final IntegerLeastSquareSolution[] recovered = new IntegerLeastSquareSolution[solutions.size()];
int k = 0;
final long[] a = new long[n];
for (final IntegerLeastSquareSolution solution : solutions) {
final long[] s = solution.getSolution();
for (int i = 0; i < n; ++i) {
// compute a = Z⁻ᵀ.s
long ai = 0;
int l = zIndex(0, i);
for (int j = 0; j < n; ++j) {
ai += zInverseTransformation[l] * s[j];
l += n;
}
a[i] = ai;
}
recovered[k++] = new IntegerLeastSquareSolution(a, solution.getSquaredDistance());
}
return recovered;
}
/** Get the index of an entry in the lower triangular matrix.
* @param row row index (counting from 0)
* @param col column index (counting from 0)
* @return index in the single dimension array
*/
protected int lIndex(final int row, final int col) {
return (row * (row - 1)) / 2 + col;
}
/** Get the index of an entry in the Z transformation matrix.
* @param row row index (counting from 0)
* @param col column index (counting from 0)
* @return index in the single dimension array
*/
protected int zIndex(final int row, final int col) {
return row * n + col;
}
}