001/*
002 * Licensed to the Apache Software Foundation (ASF) under one or more
003 * contributor license agreements.  See the NOTICE file distributed with
004 * this work for additional information regarding copyright ownership.
005 * The ASF licenses this file to You under the Apache License, Version 2.0
006 * (the "License"); you may not use this file except in compliance with
007 * the License.  You may obtain a copy of the License at
008 *
009 *      http://www.apache.org/licenses/LICENSE-2.0
010 *
011 * Unless required by applicable law or agreed to in writing, software
012 * distributed under the License is distributed on an "AS IS" BASIS,
013 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
014 * See the License for the specific language governing permissions and
015 * limitations under the License.
016 */
017package org.apache.commons.math4.transform;
018
019import java.util.function.UnaryOperator;
020import java.util.function.DoubleUnaryOperator;
021
022import org.apache.commons.numbers.complex.Complex;
023import org.apache.commons.numbers.core.ArithmeticUtils;
024
025/**
026 * Implements the Fast Cosine Transform for transformation of one-dimensional
027 * real data sets. For reference, see James S. Walker, <em>Fast Fourier
028 * Transforms</em>, chapter 3 (ISBN 0849371635).
029 * <p>
030 * There are several variants of the discrete cosine transform. The present
031 * implementation corresponds to DCT-I, with various normalization conventions,
032 * which are specified by the parameter {@link Norm}.
033 * <p>
034 * DCT-I is equivalent to DFT of an <em>even extension</em> of the data series.
035 * More precisely, if x<sub>0</sub>, &hellip;, x<sub>N-1</sub> is the data set
036 * to be cosine transformed, the extended data set
037 * x<sub>0</sub><sup>&#35;</sup>, &hellip;, x<sub>2N-3</sub><sup>&#35;</sup>
038 * is defined as follows
039 * <ul>
040 * <li>x<sub>k</sub><sup>&#35;</sup> = x<sub>k</sub> if 0 &le; k &lt; N,</li>
041 * <li>x<sub>k</sub><sup>&#35;</sup> = x<sub>2N-2-k</sub>
042 * if N &le; k &lt; 2N - 2.</li>
043 * </ul>
044 * <p>
045 * Then, the standard DCT-I y<sub>0</sub>, &hellip;, y<sub>N-1</sub> of the real
046 * data set x<sub>0</sub>, &hellip;, x<sub>N-1</sub> is equal to <em>half</em>
047 * of the N first elements of the DFT of the extended data set
048 * x<sub>0</sub><sup>&#35;</sup>, &hellip;, x<sub>2N-3</sub><sup>&#35;</sup>
049 * <br>
050 * y<sub>n</sub> = (1 / 2) &sum;<sub>k=0</sub><sup>2N-3</sup>
051 * x<sub>k</sub><sup>&#35;</sup> exp[-2&pi;i nk / (2N - 2)]
052 * &nbsp;&nbsp;&nbsp;&nbsp;k = 0, &hellip;, N-1.
053 * <p>
054 * The present implementation of the discrete cosine transform as a fast cosine
055 * transform requires the length of the data set to be a power of two plus one
056 * (N&nbsp;=&nbsp;2<sup>n</sup>&nbsp;+&nbsp;1). Besides, it implicitly assumes
057 * that the sampled function is even.
058 */
059public class FastCosineTransform implements RealTransform {
060    /** Operation to be performed. */
061    private final UnaryOperator<double[]> op;
062
063    /**
064     * @param normalization Normalization to be applied to the
065     * transformed data.
066     * @param inverse Whether to perform the inverse transform.
067     */
068    public FastCosineTransform(final Norm normalization,
069                               final boolean inverse) {
070        op = create(normalization, inverse);
071    }
072
073    /**
074     * @param normalization Normalization to be applied to the
075     * transformed data.
076     */
077    public FastCosineTransform(final Norm normalization) {
078        this(normalization, false);
079    }
080
081    /**
082     * {@inheritDoc}
083     *
084     * @throws IllegalArgumentException if the length of the data array is
085     * not a power of two plus one.
086     */
087    @Override
088    public double[] apply(final double[] f) {
089        return op.apply(f);
090    }
091
092    /**
093     * {@inheritDoc}
094     *
095     * @throws IllegalArgumentException if the number of sample points is
096     * not a power of two plus one, if the lower bound is greater than or
097     * equal to the upper bound, if the number of sample points is negative.
098     */
099    @Override
100    public double[] apply(final DoubleUnaryOperator f,
101                          final double min,
102                          final double max,
103                          final int n) {
104        return apply(TransformUtils.sample(f, min, max, n));
105    }
106
107    /**
108     * Perform the FCT algorithm (including inverse).
109     *
110     * @param f Data to be transformed.
111     * @return the transformed array.
112     * @throws IllegalArgumentException if the length of the data array is
113     * not a power of two plus one.
114     */
115    private double[] fct(double[] f) {
116        final int n = f.length - 1;
117        if (!ArithmeticUtils.isPowerOfTwo(n)) {
118            throw new TransformException(TransformException.NOT_POWER_OF_TWO_PLUS_ONE,
119                                         Integer.valueOf(f.length));
120        }
121
122        final double[] transformed = new double[f.length];
123
124        if (n == 1) {       // trivial case
125            transformed[0] = 0.5 * (f[0] + f[1]);
126            transformed[1] = 0.5 * (f[0] - f[1]);
127            return transformed;
128        }
129
130        // construct a new array and perform FFT on it
131        final double[] x = new double[n];
132        x[0] = 0.5 * (f[0] + f[n]);
133        final int nShifted = n >> 1;
134        x[nShifted] = f[nShifted];
135        // temporary variable for transformed[1]
136        double t1 = 0.5 * (f[0] - f[n]);
137        final double piOverN = Math.PI / n;
138        for (int i = 1; i < nShifted; i++) {
139            final int nMi = n - i;
140            final double fi = f[i];
141            final double fnMi = f[nMi];
142            final double a = 0.5 * (fi + fnMi);
143            final double arg = i * piOverN;
144            final double b = Math.sin(arg) * (fi - fnMi);
145            final double c = Math.cos(arg) * (fi - fnMi);
146            x[i] = a - b;
147            x[nMi] = a + b;
148            t1 += c;
149        }
150        final FastFourierTransform transformer = new FastFourierTransform(FastFourierTransform.Norm.STD,
151                                                                          false);
152        final Complex[] y = transformer.apply(x);
153
154        // reconstruct the FCT result for the original array
155        transformed[0] = y[0].getReal();
156        transformed[1] = t1;
157        for (int i = 1; i < nShifted; i++) {
158            final int i2 = 2 * i;
159            transformed[i2] = y[i].getReal();
160            transformed[i2 + 1] = transformed[i2 - 1] - y[i].getImaginary();
161        }
162        transformed[n] = y[nShifted].getReal();
163
164        return transformed;
165    }
166
167    /**
168     * Factory method.
169     *
170     * @param normalization Normalization to be applied to the
171     * transformed data.
172     * @param inverse Whether to perform the inverse transform.
173     * @return the transform operator.
174     */
175    private UnaryOperator<double[]> create(final Norm normalization,
176                                           final boolean inverse) {
177        if (inverse) {
178            return normalization == Norm.ORTHO ?
179                f -> TransformUtils.scaleInPlace(fct(f), Math.sqrt(2d / (f.length - 1))) :
180                f -> TransformUtils.scaleInPlace(fct(f), 2d / (f.length - 1));
181        } else {
182            return normalization == Norm.ORTHO ?
183                f -> TransformUtils.scaleInPlace(fct(f), Math.sqrt(2d / (f.length - 1))) :
184                f -> fct(f);
185        }
186    }
187
188    /**
189     * Normalization types.
190     */
191    public enum Norm {
192        /**
193         * Should be passed to the constructor of {@link FastCosineTransform}
194         * to use the <em>standard</em> normalization convention.  The standard
195         * DCT-I normalization convention is defined as follows
196         * <ul>
197         * <li>forward transform:
198         * y<sub>n</sub> = (1/2) [x<sub>0</sub> + (-1)<sup>n</sup>x<sub>N-1</sub>]
199         * + &sum;<sub>k=1</sub><sup>N-2</sup>
200         * x<sub>k</sub> cos[&pi; nk / (N - 1)],</li>
201         * <li>inverse transform:
202         * x<sub>k</sub> = [1 / (N - 1)] [y<sub>0</sub>
203         * + (-1)<sup>k</sup>y<sub>N-1</sub>]
204         * + [2 / (N - 1)] &sum;<sub>n=1</sub><sup>N-2</sup>
205         * y<sub>n</sub> cos[&pi; nk / (N - 1)],</li>
206         * </ul>
207         * where N is the size of the data sample.
208         */
209        STD,
210
211        /**
212         * Should be passed to the constructor of {@link FastCosineTransform}
213         * to use the <em>orthogonal</em> normalization convention. The orthogonal
214         * DCT-I normalization convention is defined as follows
215         * <ul>
216         * <li>forward transform:
217         * y<sub>n</sub> = [2(N - 1)]<sup>-1/2</sup> [x<sub>0</sub>
218         * + (-1)<sup>n</sup>x<sub>N-1</sub>]
219         * + [2 / (N - 1)]<sup>1/2</sup> &sum;<sub>k=1</sub><sup>N-2</sup>
220         * x<sub>k</sub> cos[&pi; nk / (N - 1)],</li>
221         * <li>inverse transform:
222         * x<sub>k</sub> = [2(N - 1)]<sup>-1/2</sup> [y<sub>0</sub>
223         * + (-1)<sup>k</sup>y<sub>N-1</sub>]
224         * + [2 / (N - 1)]<sup>1/2</sup> &sum;<sub>n=1</sub><sup>N-2</sup>
225         * y<sub>n</sub> cos[&pi; nk / (N - 1)],</li>
226         * </ul>
227         * which makes the transform orthogonal. N is the size of the data sample.
228         */
229        ORTHO;
230    }
231}