1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

61

62

63

64

65

66

67

68

69

70

71

72

73

74

75

76

77

78

79

80

81

82

83

84

85

86

87

88

89

90

91

92

93

94

95

96

97

98

99

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

403

404

405

406

407

408

409

410

411

412

413

414

415

416

417

418

419

420

421

422

423

424

425

426

427

428

429

430

431

432

433

434

435

436

437

438

439

440

441

442

443

444

445

446

447

448

449

450

451

452

453

454

455

456

457

458

459

460

461

462

463

464

465

466

467

468

469

470

471

472

473

474

475

476

477

478

479

480

481

482

483

484

485

486

487

488

489

490

491

492

493

494

495

496

497

498

499

500

501

502

503

504

505

506

507

508

509

510

511

512

513

514

515

516

517

518

519

520

521

522

523

524

525

526

527

528

529

530

531

532

533

534

535

536

537

538

539

540

541

542

543

544

545

546

547

548

549

550

551

552

553

554

555

556

557

558

559

560

561

562

563

564

565

566

567

568

569

570

571

572

573

574

575

576

577

578

579

580

581

582

583

584

585

586

587

588

589

590

591

592

593

594

595

596

597

598

599

600

601

602

603

604

605

606

607

608

609

610

611

612

613

614

615

616

617

618

619

620

621

622

623

624

625

626

627

628

629

630

631

632

633

634

635

636

637

638

639

640

641

642

643

644

645

646

647

648

649

650

651

652

653

654

655

656

657

658

659

660

661

662

663

664

665

666

667

668

669

670

671

672

673

674

675

676

677

678

679

680

681

682

683

684

685

686

687

688

689

690

691

692

693

694

695

696

697

698

699

700

701

702

703

704

705

706

707

708

709

710

711

712

713

714

715

716

717

718

719

720

721

722

723

724

725

726

727

728

729

730

731

732

733

734

735

736

737

738

739

740

741

742

743

744

745

746

747

748

749

750

751

752

753

754

755

756

757

758

759

760

761

762

763

764

765

766

767

768

769

770

771

772

773

774

775

776

777

778

779

780

781

782

783

784

785

786

787

788

789

790

791

792

793

794

795

796

797

798

799

800

801

802

803

804

805

806

807

808

809

810

811

812

813

814

815

816

817

818

819

820

821

822

823

824

825

826

827

828

829

830

831

832

833

834

835

836

837

838

839

840

841

842

843

844

845

846

847

848

849

850

851

852

853

854

855

856

857

858

859

860

861

862

863

864

865

866

867

868

869

870

871

872

873

874

875

876

877

878

879

880

881

882

883

884

885

886

887

888

889

890

891

892

893

894

895

896

897

898

899

900

901

902

903

904

905

906

907

908

909

910

911

912

913

914

915

916

917

918

919

920

921

922

923

924

925

926

927

928

929

930

931

932

933

934

935

936

937

938

939

940

941

942

943

944

945

946

947

948

949

950

951

952

953

954

955

956

957

958

959

960

961

962

963

964

965

966

967

968

969

970

971

972

973

974

975

976

977

978

979

980

981

982

983

984

985

986

987

988

989

990

991

992

993

994

995

996

997

998

999

1000

1001

1002

1003

1004

1005

1006

1007

1008

1009

1010

1011

1012

1013

1014

1015

1016

1017

1018

1019

1020

1021

1022

1023

1024

1025

1026

1027

1028

1029

1030

1031

1032

1033

1034

1035

1036

1037

1038

1039

1040

1041

1042

1043

1044

1045

1046

1047

1048

1049

1050

1051

1052

1053

1054

1055

1056

1057

1058

1059

1060

1061

1062

1063

1064

1065

1066

1067

1068

1069

1070

1071

1072

1073

1074

1075

1076

1077

1078

1079

1080

1081

1082

1083

1084

1085

1086

1087

1088

1089

1090

1091

1092

1093

1094

1095

1096

1097

1098

1099

1100

1101

1102

1103

1104

1105

1106

1107

1108

1109

1110

1111

1112

1113

1114

1115

1116

1117

1118

1119

1120

1121

1122

1123

1124

1125

1126

1127

1128

1129

1130

1131

1132

1133

1134

1135

1136

1137

1138

1139

1140

1141

1142

1143

1144

1145

1146

1147

1148

1149

1150

1151

1152

1153

1154

1155

1156

1157

1158

1159

1160

1161

1162

1163

1164

1165

1166

1167

1168

1169

1170

1171

1172

1173

1174

1175

1176

1177

1178

1179

1180

1181

1182

1183

1184

1185

1186

1187

1188

1189

1190

1191

1192

1193

1194

1195

1196

1197

1198

1199

1200

1201

1202

1203

1204

1205

1206

1207

1208

1209

1210

1211

1212

1213

1214

1215

1216

1217

1218

1219

1220

1221

1222

1223

1224

1225

1226

1227

1228

1229

1230

1231

1232

1233

1234

1235

1236

1237

1238

1239

1240

1241

1242

1243

1244

1245

1246

1247

1248

1249

1250

1251

1252

1253

1254

1255

1256

1257

1258

1259

1260

1261

1262

1263

1264

1265

1266

1267

1268

1269

1270

1271

1272

1273

1274

1275

1276

1277

1278

1279

1280

1281

1282

1283

1284

1285

1286

1287

1288

1289

1290

1291

1292

1293

1294

1295

1296

1297

1298

1299

1300

1301

1302

1303

1304

1305

1306

1307

1308

1309

1310

1311

1312

1313

1314

1315

1316

1317

1318

1319

1320

1321

1322

1323

1324

1325

1326

1327

1328

1329

1330

1331

1332

1333

1334

1335

1336

1337

1338

1339

1340

1341

1342

1343

1344

1345

1346

1347

1348

1349

1350

1351

1352

1353

1354

1355

1356

1357

1358

1359

1360

1361

1362

1363

1364

1365

1366

1367

1368

1369

1370

1371

1372

1373

1374

1375

1376

1377

1378

1379

1380

1381

1382

1383

1384

1385

1386

1387

1388

1389

1390

1391

1392

1393

1394

1395

1396

1397

1398

1399

1400

1401

1402

1403

1404

1405

1406

1407

1408

1409

1410

1411

1412

1413

1414

1415

1416

1417

1418

1419

1420

1421

1422

1423

1424

1425

1426

1427

1428

1429

1430

1431

1432

1433

1434

1435

1436

1437

1438

1439

1440

1441

1442

1443

1444

1445

1446

1447

1448

1449

1450

1451

1452

1453

1454

1455

1456

1457

1458

1459

1460

1461

1462

1463

1464

1465

1466

1467

1468

1469

1470

1471

1472

1473

1474

1475

1476

1477

1478

1479

1480

1481

1482

1483

1484

1485

1486

1487

1488

1489

1490

1491

1492

1493

1494

1495

1496

1497

1498

1499

1500

1501

1502

1503

1504

1505

1506

1507

1508

1509

1510

1511

1512

1513

1514

1515

1516

1517

1518

1519

1520

1521

1522

1523

1524

1525

1526

1527

1528

1529

1530

1531

1532

1533

1534

1535

1536

1537

1538

1539

1540

1541

1542

1543

1544

1545

1546

1547

1548

1549

1550

1551

1552

1553

1554

1555

1556

1557

1558

1559

1560

1561

1562

1563

1564

1565

1566

1567

1568

1569

1570

1571

1572

1573

1574

1575

1576

1577

1578

1579

1580

1581

1582

1583

1584

1585

1586

1587

1588

1589

1590

1591

1592

1593

1594

1595

1596

1597

1598

1599

1600

1601

1602

1603

1604

1605

1606

1607

1608

1609

1610

1611

1612

1613

1614

1615

1616

1617

1618

1619

1620

1621

1622

1623

1624

1625

1626

1627

1628

1629

1630

1631

1632

1633

1634

1635

1636

1637

1638

1639

1640

1641

1642

1643

1644

1645

1646

1647

1648

1649

1650

1651

1652

1653

1654

1655

1656

1657

1658

1659

1660

1661

1662

1663

1664

1665

1666

1667

1668

1669

1670

1671

1672

1673

1674

1675

1676

1677

1678

1679

1680

1681

1682

1683

1684

1685

1686

1687

1688

1689

1690

1691

1692

1693

1694

1695

1696

1697

1698

1699

1700

1701

1702

1703

1704

1705

1706

1707

1708

1709

1710

1711

1712

1713

1714

1715

1716

1717

1718

1719

1720

1721

1722

1723

1724

1725

1726

1727

1728

1729

1730

1731

1732

1733

1734

1735

1736

1737

1738

1739

1740

1741

1742

1743

1744

1745

1746

1747

1748

1749

1750

1751

1752

1753

1754

1755

1756

1757

1758

1759

1760

1761

1762

1763

1764

1765

1766

1767

1768

1769

1770

1771

1772

1773

1774

1775

1776

1777

1778

1779

1780

1781

1782

1783

1784

1785

1786

1787

1788

1789

1790

1791

1792

1793

1794

1795

1796

1797

1798

1799

1800

1801

1802

1803

1804

1805

1806

1807

1808

1809

1810

1811

1812

1813

1814

1815

1816

1817

1818

1819

1820

1821

1822

1823

1824

1825

1826

1827

1828

1829

1830

1831

1832

1833

1834

1835

1836

1837

1838

1839

1840

1841

1842

1843

1844

1845

1846

1847

1848

1849

1850

1851

1852

1853

1854

1855

1856

1857

1858

1859

1860

1861

1862

1863

1864

1865

1866

1867

1868

1869

1870

1871

1872

1873

1874

1875

1876

1877

1878

1879

1880

1881

1882

1883

1884

1885

1886

1887

1888

1889

1890

1891

1892

1893

1894

1895

1896

1897

1898

1899

1900

1901

1902

1903

1904

1905

1906

1907

1908

1909

1910

1911

1912

1913

1914

1915

1916

1917

1918

1919

1920

1921

1922

1923

1924

1925

1926

1927

1928

1929

1930

1931

1932

1933

1934

1935

1936

1937

1938

1939

1940

1941

1942

1943

1944

1945

1946

1947

1948

1949

1950

1951

1952

1953

1954

1955

1956

1957

1958

1959

1960

1961

1962

1963

1964

1965

1966

1967

1968

1969

1970

1971

1972

1973

1974

1975

1976

1977

1978

1979

1980

1981

1982

1983

1984

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

2017

2018

2019

2020

2021

2022

2023

2024

2025

2026

2027

2028

2029

2030

2031

2032

2033

2034

2035

2036

2037

2038

2039

2040

2041

2042

2043

2044

2045

2046

2047

2048

2049

2050

2051

2052

2053

2054

2055

2056

2057

2058

2059

2060

2061

2062

2063

2064

2065

2066

2067

2068

2069

2070

2071

2072

2073

2074

2075

2076

2077

2078

2079

2080

2081

2082

2083

2084

2085

2086

2087

2088

2089

2090

2091

2092

2093

2094

2095

2096

2097

2098

2099

2100

2101

2102

2103

2104

2105

2106

2107

2108

2109

2110

2111

2112

2113

2114

2115

2116

2117

2118

2119

2120

2121

2122

2123

2124

2125

2126

2127

2128

2129

2130

2131

2132

2133

2134

2135

2136

2137

2138

2139

2140

2141

2142

2143

2144

2145

2146

2147

2148

2149

2150

2151

2152

2153

2154

2155

2156

2157

2158

2159

2160

2161

2162

2163

2164

2165

2166

2167

2168

2169

2170

2171

2172

2173

2174

2175

2176

2177

2178

2179

2180

2181

2182

2183

2184

2185

2186

2187

2188

2189

2190

2191

2192

2193

2194

2195

2196

2197

2198

2199

2200

2201

2202

2203

2204

2205

2206

2207

2208

2209

2210

2211

2212

2213

2214

2215

2216

2217

2218

2219

2220

2221

2222

2223

2224

2225

2226

2227

2228

2229

2230

2231

2232

2233

2234

2235

2236

2237

2238

2239

2240

2241

2242

2243

2244

2245

2246

2247

2248

2249

2250

2251

2252

2253

2254

2255

2256

2257

2258

2259

2260

2261

2262

2263

2264

2265

2266

2267

2268

2269

2270

2271

2272

2273

2274

2275

2276

2277

2278

2279

2280

2281

2282

2283

2284

2285

2286

2287

2288

2289

2290

2291

2292

2293

2294

2295

2296

2297

2298

2299

2300

2301

2302

2303

2304

2305

2306

2307

2308

2309

2310

2311

2312

2313

2314

2315

2316

2317

2318

2319

2320

2321

2322

2323

2324

2325

2326

2327

2328

2329

2330

2331

2332

2333

2334

2335

2336

2337

2338

2339

2340

2341

2342

2343

2344

2345

2346

2347

2348

2349

2350

2351

2352

2353

2354

2355

2356

2357

2358

2359

2360

2361

2362

2363

2364

2365

2366

2367

2368

2369

2370

2371

2372

2373

2374

2375

2376

2377

2378

2379

2380

2381

2382

2383

2384

2385

2386

2387

2388

2389

2390

2391

2392

2393

2394

2395

2396

2397

2398

2399

2400

2401

2402

2403

2404

2405

2406

2407

2408

2409

2410

2411

2412

2413

2414

2415

2416

2417

2418

2419

2420

2421

2422

2423

2424

2425

2426

2427

2428

2429

2430

2431

2432

2433

2434

2435

2436

2437

2438

2439

2440

2441

2442

2443

2444

2445

2446

2447

2448

2449

2450

2451

2452

2453

2454

2455

2456

2457

2458

2459

2460

2461

2462

2463

2464

2465

2466

2467

2468

2469

2470

2471

2472

2473

2474

2475

2476

2477

2478

2479

2480

2481

2482

2483

2484

2485

2486

2487

2488

2489

2490

2491

2492

2493

2494

2495

2496

2497

2498

2499

2500

2501

2502

2503

2504

2505

2506

2507

2508

2509

2510

2511

2512

2513

2514

2515

2516

2517

2518

2519

2520

2521

2522

2523

2524

2525

2526

2527

2528

2529

2530

2531

2532

2533

2534

2535

2536

2537

2538

2539

2540

2541

2542

2543

2544

2545

2546

2547

2548

2549

2550

2551

2552

2553

2554

2555

2556

2557

2558

2559

2560

2561

2562

2563

2564

2565

2566

2567

2568

2569

2570

2571

2572

2573

2574

2575

2576

2577

2578

2579

2580

2581

2582

2583

2584

2585

2586

2587

2588

2589

2590

2591

2592

2593

2594

2595

2596

2597

2598

2599

2600

2601

2602

2603

2604

2605

2606

2607

2608

2609

2610

2611

2612

2613

2614

2615

2616

2617

2618

2619

2620

2621

2622

2623

2624

2625

2626

2627

2628

2629

2630

2631

2632

2633

2634

2635

2636

2637

2638

2639

2640

2641

2642

2643

2644

2645

2646

2647

2648

2649

2650

2651

2652

2653

2654

2655

2656

2657

2658

2659

2660

2661

2662

2663

2664

2665

2666

2667

2668

2669

2670

2671

2672

2673

2674

2675

2676

2677

2678

2679

2680

2681

2682

2683

2684

2685

2686

2687

2688

2689

2690

2691

2692

2693

2694

2695

2696

2697

2698

2699

2700

2701

2702

2703

2704

2705

2706

2707

2708

2709

2710

2711

2712

2713

2714

2715

2716

2717

2718

2719

2720

2721

2722

2723

2724

2725

2726

2727

2728

2729

2730

2731

2732

2733

2734

2735

2736

2737

2738

2739

2740

2741

2742

2743

2744

2745

2746

2747

2748

2749

2750

2751

2752

2753

2754

2755

2756

2757

2758

2759

2760

2761

2762

2763

2764

2765

2766

2767

2768

2769

2770

2771

2772

2773

2774

2775

2776

2777

2778

2779

2780

2781

2782

2783

2784

2785

2786

2787

2788

2789

2790

2791

2792

2793

2794

2795

2796

2797

2798

2799

2800

2801

2802

2803

2804

2805

2806

2807

2808

2809

2810

2811

2812

2813

2814

2815

2816

2817

2818

2819

2820

2821

2822

2823

2824

2825

2826

2827

2828

2829

2830

2831

2832

2833

2834

2835

2836

2837

2838

2839

2840

2841

2842

2843

2844

2845

2846

2847

2848

2849

2850

2851

2852

2853

2854

2855

2856

2857

2858

2859

2860

2861

2862

2863

2864

2865

2866

2867

2868

2869

2870

2871

2872

2873

2874

2875

2876

2877

2878

2879

2880

2881

2882

2883

2884

2885

2886

2887

2888

2889

2890

2891

2892

2893

2894

2895

2896

2897

2898

2899

2900

2901

2902

2903

2904

2905

2906

2907

2908

2909

2910

2911

2912

2913

2914

2915

2916

2917

2918

2919

2920

2921

2922

2923

2924

2925

2926

2927

2928

2929

2930

2931

2932

2933

2934

2935

2936

2937

2938

2939

2940

2941

2942

2943

2944

2945

2946

2947

2948

2949

2950

2951

2952

2953

2954

2955

2956

2957

2958

2959

2960

2961

2962

2963

2964

2965

2966

2967

2968

2969

2970

2971

2972

2973

2974

2975

2976

2977

2978

2979

2980

2981

2982

2983

2984

2985

2986

2987

2988

2989

2990

2991

2992

2993

2994

2995

2996

2997

2998

2999

3000

3001

3002

3003

3004

3005

3006

3007

3008

3009

3010

3011

3012

3013

3014

3015

3016

3017

3018

3019

3020

3021

3022

3023

3024

3025

3026

3027

3028

3029

3030

3031

3032

3033

3034

3035

3036

3037

3038

3039

3040

3041

3042

3043

3044

3045

3046

3047

3048

3049

3050

3051

3052

3053

3054

3055

3056

3057

3058

3059

3060

3061

3062

3063

3064

3065

3066

3067

3068

3069

3070

3071

3072

3073

3074

3075

3076

3077

3078

3079

3080

3081

3082

3083

3084

3085

3086

3087

3088

3089

3090

3091

3092

3093

3094

3095

3096

3097

3098

3099

3100

3101

3102

3103

3104

3105

3106

3107

3108

3109

3110

3111

3112

3113

3114

3115

3116

3117

3118

3119

3120

3121

3122

3123

3124

3125

3126

3127

3128

3129

3130

3131

3132

3133

3134

3135

3136

3137

3138

3139

3140

3141

3142

3143

3144

3145

3146

3147

3148

3149

3150

3151

3152

3153

3154

3155

3156

3157

3158

3159

3160

3161

3162

3163

3164

3165

3166

3167

3168

3169

3170

3171

3172

3173

3174

3175

3176

3177

3178

3179

3180

3181

3182

3183

3184

3185

3186

3187

3188

3189

3190

3191

3192

3193

3194

3195

3196

3197

3198

3199

3200

3201

3202

3203

3204

3205

3206

3207

3208

3209

3210

3211

3212

3213

3214

3215

3216

3217

3218

3219

3220

3221

3222

3223

3224

3225

3226

3227

3228

3229

3230

3231

3232

3233

3234

3235

3236

3237

3238

3239

3240

3241

3242

3243

3244

3245

3246

3247

3248

3249

3250

3251

3252

3253

3254

3255

3256

3257

3258

3259

3260

3261

3262

3263

3264

3265

3266

3267

3268

3269

3270

3271

3272

3273

3274

3275

3276

3277

3278

3279

3280

3281

3282

3283

3284

3285

3286

3287

3288

3289

3290

3291

3292

3293

3294

3295

3296

3297

3298

3299

3300

3301

3302

3303

3304

3305

3306

3307

3308

3309

3310

3311

3312

3313

3314

3315

3316

3317

3318

3319

3320

3321

3322

3323

3324

3325

3326

3327

3328

3329

3330

3331

3332

3333

3334

3335

3336

3337

3338

3339

3340

3341

3342

3343

3344

3345

3346

3347

3348

3349

3350

3351

3352

3353

3354

3355

3356

3357

3358

3359

3360

3361

3362

3363

3364

3365

3366

3367

3368

3369

3370

3371

3372

3373

3374

3375

3376

3377

3378

3379

3380

3381

3382

3383

3384

3385

3386

3387

3388

3389

3390

3391

3392

3393

3394

3395

3396

3397

3398

3399

3400

3401

3402

3403

3404

3405

3406

3407

3408

3409

3410

3411

3412

3413

3414

3415

3416

3417

3418

3419

3420

3421

3422

3423

3424

3425

3426

3427

3428

3429

3430

3431

3432

3433

3434

3435

3436

3437

3438

3439

3440

3441

3442

3443

3444

3445

3446

3447

3448

3449

3450

3451

3452

3453

3454

3455

3456

3457

3458

3459

3460

3461

3462

3463

# Author: Travis Oliphant 

# 1999 -- 2002 

 

from __future__ import division, print_function, absolute_import 

 

import operator 

import threading 

import sys 

import timeit 

 

from . import sigtools, dlti 

from ._upfirdn import upfirdn, _output_len 

from scipy._lib.six import callable 

from scipy._lib._version import NumpyVersion 

from scipy import fftpack, linalg 

from numpy import (allclose, angle, arange, argsort, array, asarray, 

atleast_1d, atleast_2d, cast, dot, exp, expand_dims, 

iscomplexobj, mean, ndarray, newaxis, ones, pi, 

poly, polyadd, polyder, polydiv, polymul, polysub, polyval, 

product, r_, ravel, real_if_close, reshape, 

roots, sort, take, transpose, unique, where, zeros, 

zeros_like) 

import numpy as np 

import math 

from scipy.special import factorial 

from .windows import get_window 

from ._arraytools import axis_slice, axis_reverse, odd_ext, even_ext, const_ext 

from .filter_design import cheby1, _validate_sos 

from .fir_filter_design import firwin 

 

if sys.version_info.major >= 3 and sys.version_info.minor >= 5: 

from math import gcd 

else: 

from fractions import gcd 

 

 

__all__ = ['correlate', 'fftconvolve', 'convolve', 'convolve2d', 'correlate2d', 

'order_filter', 'medfilt', 'medfilt2d', 'wiener', 'lfilter', 

'lfiltic', 'sosfilt', 'deconvolve', 'hilbert', 'hilbert2', 

'cmplx_sort', 'unique_roots', 'invres', 'invresz', 'residue', 

'residuez', 'resample', 'resample_poly', 'detrend', 

'lfilter_zi', 'sosfilt_zi', 'sosfiltfilt', 'choose_conv_method', 

'filtfilt', 'decimate', 'vectorstrength'] 

 

 

_modedict = {'valid': 0, 'same': 1, 'full': 2} 

 

_boundarydict = {'fill': 0, 'pad': 0, 'wrap': 2, 'circular': 2, 'symm': 1, 

'symmetric': 1, 'reflect': 4} 

 

 

_rfft_mt_safe = (NumpyVersion(np.__version__) >= '1.9.0.dev-e24486e') 

 

_rfft_lock = threading.Lock() 

 

 

def _valfrommode(mode): 

try: 

return _modedict[mode] 

except KeyError: 

raise ValueError("Acceptable mode flags are 'valid'," 

" 'same', or 'full'.") 

 

 

def _bvalfromboundary(boundary): 

try: 

return _boundarydict[boundary] << 2 

except KeyError: 

raise ValueError("Acceptable boundary flags are 'fill', 'circular' " 

"(or 'wrap'), and 'symmetric' (or 'symm').") 

 

 

def _inputs_swap_needed(mode, shape1, shape2): 

""" 

If in 'valid' mode, returns whether or not the input arrays need to be 

swapped depending on whether `shape1` is at least as large as `shape2` in 

every dimension. 

 

This is important for some of the correlation and convolution 

implementations in this module, where the larger array input needs to come 

before the smaller array input when operating in this mode. 

 

Note that if the mode provided is not 'valid', False is immediately 

returned. 

""" 

if mode == 'valid': 

ok1, ok2 = True, True 

 

for d1, d2 in zip(shape1, shape2): 

if not d1 >= d2: 

ok1 = False 

if not d2 >= d1: 

ok2 = False 

 

if not (ok1 or ok2): 

raise ValueError("For 'valid' mode, one must be at least " 

"as large as the other in every dimension") 

 

return not ok1 

 

return False 

 

 

def correlate(in1, in2, mode='full', method='auto'): 

r""" 

Cross-correlate two N-dimensional arrays. 

 

Cross-correlate `in1` and `in2`, with the output size determined by the 

`mode` argument. 

 

Parameters 

---------- 

in1 : array_like 

First input. 

in2 : array_like 

Second input. Should have the same number of dimensions as `in1`. 

mode : str {'full', 'valid', 'same'}, optional 

A string indicating the size of the output: 

 

``full`` 

The output is the full discrete linear cross-correlation 

of the inputs. (Default) 

``valid`` 

The output consists only of those elements that do not 

rely on the zero-padding. In 'valid' mode, either `in1` or `in2` 

must be at least as large as the other in every dimension. 

``same`` 

The output is the same size as `in1`, centered 

with respect to the 'full' output. 

method : str {'auto', 'direct', 'fft'}, optional 

A string indicating which method to use to calculate the correlation. 

 

``direct`` 

The correlation is determined directly from sums, the definition of 

correlation. 

``fft`` 

The Fast Fourier Transform is used to perform the correlation more 

quickly (only available for numerical arrays.) 

``auto`` 

Automatically chooses direct or Fourier method based on an estimate 

of which is faster (default). See `convolve` Notes for more detail. 

 

.. versionadded:: 0.19.0 

 

Returns 

------- 

correlate : array 

An N-dimensional array containing a subset of the discrete linear 

cross-correlation of `in1` with `in2`. 

 

See Also 

-------- 

choose_conv_method : contains more documentation on `method`. 

 

Notes 

----- 

The correlation z of two d-dimensional arrays x and y is defined as:: 

 

z[...,k,...] = sum[..., i_l, ...] x[..., i_l,...] * conj(y[..., i_l - k,...]) 

 

This way, if x and y are 1-D arrays and ``z = correlate(x, y, 'full')`` 

then 

 

.. math:: 

 

z[k] = (x * y)(k - N + 1) 

= \sum_{l=0}^{||x||-1}x_l y_{l-k+N-1}^{*} 

 

for :math:`k = 0, 1, ..., ||x|| + ||y|| - 2` 

 

where :math:`||x||` is the length of ``x``, :math:`N = \max(||x||,||y||)`, 

and :math:`y_m` is 0 when m is outside the range of y. 

 

``method='fft'`` only works for numerical arrays as it relies on 

`fftconvolve`. In certain cases (i.e., arrays of objects or when 

rounding integers can lose precision), ``method='direct'`` is always used. 

 

Examples 

-------- 

Implement a matched filter using cross-correlation, to recover a signal 

that has passed through a noisy channel. 

 

>>> from scipy import signal 

>>> sig = np.repeat([0., 1., 1., 0., 1., 0., 0., 1.], 128) 

>>> sig_noise = sig + np.random.randn(len(sig)) 

>>> corr = signal.correlate(sig_noise, np.ones(128), mode='same') / 128 

 

>>> import matplotlib.pyplot as plt 

>>> clock = np.arange(64, len(sig), 128) 

>>> fig, (ax_orig, ax_noise, ax_corr) = plt.subplots(3, 1, sharex=True) 

>>> ax_orig.plot(sig) 

>>> ax_orig.plot(clock, sig[clock], 'ro') 

>>> ax_orig.set_title('Original signal') 

>>> ax_noise.plot(sig_noise) 

>>> ax_noise.set_title('Signal with noise') 

>>> ax_corr.plot(corr) 

>>> ax_corr.plot(clock, corr[clock], 'ro') 

>>> ax_corr.axhline(0.5, ls=':') 

>>> ax_corr.set_title('Cross-correlated with rectangular pulse') 

>>> ax_orig.margins(0, 0.1) 

>>> fig.tight_layout() 

>>> fig.show() 

 

""" 

in1 = asarray(in1) 

in2 = asarray(in2) 

 

if in1.ndim == in2.ndim == 0: 

return in1 * in2.conj() 

elif in1.ndim != in2.ndim: 

raise ValueError("in1 and in2 should have the same dimensionality") 

 

# Don't use _valfrommode, since correlate should not accept numeric modes 

try: 

val = _modedict[mode] 

except KeyError: 

raise ValueError("Acceptable mode flags are 'valid'," 

" 'same', or 'full'.") 

 

# this either calls fftconvolve or this function with method=='direct' 

if method in ('fft', 'auto'): 

return convolve(in1, _reverse_and_conj(in2), mode, method) 

 

elif method == 'direct': 

# fastpath to faster numpy.correlate for 1d inputs when possible 

if _np_conv_ok(in1, in2, mode): 

return np.correlate(in1, in2, mode) 

 

# _correlateND is far slower when in2.size > in1.size, so swap them 

# and then undo the effect afterward if mode == 'full'. Also, it fails 

# with 'valid' mode if in2 is larger than in1, so swap those, too. 

# Don't swap inputs for 'same' mode, since shape of in1 matters. 

swapped_inputs = ((mode == 'full') and (in2.size > in1.size) or 

_inputs_swap_needed(mode, in1.shape, in2.shape)) 

 

if swapped_inputs: 

in1, in2 = in2, in1 

 

if mode == 'valid': 

ps = [i - j + 1 for i, j in zip(in1.shape, in2.shape)] 

out = np.empty(ps, in1.dtype) 

 

z = sigtools._correlateND(in1, in2, out, val) 

 

else: 

ps = [i + j - 1 for i, j in zip(in1.shape, in2.shape)] 

 

# zero pad input 

in1zpadded = np.zeros(ps, in1.dtype) 

sc = [slice(0, i) for i in in1.shape] 

in1zpadded[sc] = in1.copy() 

 

if mode == 'full': 

out = np.empty(ps, in1.dtype) 

elif mode == 'same': 

out = np.empty(in1.shape, in1.dtype) 

 

z = sigtools._correlateND(in1zpadded, in2, out, val) 

 

if swapped_inputs: 

# Reverse and conjugate to undo the effect of swapping inputs 

z = _reverse_and_conj(z) 

 

return z 

 

else: 

raise ValueError("Acceptable method flags are 'auto'," 

" 'direct', or 'fft'.") 

 

 

def _centered(arr, newshape): 

# Return the center newshape portion of the array. 

newshape = asarray(newshape) 

currshape = array(arr.shape) 

startind = (currshape - newshape) // 2 

endind = startind + newshape 

myslice = [slice(startind[k], endind[k]) for k in range(len(endind))] 

return arr[tuple(myslice)] 

 

 

def fftconvolve(in1, in2, mode="full"): 

"""Convolve two N-dimensional arrays using FFT. 

 

Convolve `in1` and `in2` using the fast Fourier transform method, with 

the output size determined by the `mode` argument. 

 

This is generally much faster than `convolve` for large arrays (n > ~500), 

but can be slower when only a few output values are needed, and can only 

output float arrays (int or object array inputs will be cast to float). 

 

As of v0.19, `convolve` automatically chooses this method or the direct 

method based on an estimation of which is faster. 

 

Parameters 

---------- 

in1 : array_like 

First input. 

in2 : array_like 

Second input. Should have the same number of dimensions as `in1`. 

mode : str {'full', 'valid', 'same'}, optional 

A string indicating the size of the output: 

 

``full`` 

The output is the full discrete linear convolution 

of the inputs. (Default) 

``valid`` 

The output consists only of those elements that do not 

rely on the zero-padding. In 'valid' mode, either `in1` or `in2` 

must be at least as large as the other in every dimension. 

``same`` 

The output is the same size as `in1`, centered 

with respect to the 'full' output. 

 

Returns 

------- 

out : array 

An N-dimensional array containing a subset of the discrete linear 

convolution of `in1` with `in2`. 

 

Examples 

-------- 

Autocorrelation of white noise is an impulse. 

 

>>> from scipy import signal 

>>> sig = np.random.randn(1000) 

>>> autocorr = signal.fftconvolve(sig, sig[::-1], mode='full') 

 

>>> import matplotlib.pyplot as plt 

>>> fig, (ax_orig, ax_mag) = plt.subplots(2, 1) 

>>> ax_orig.plot(sig) 

>>> ax_orig.set_title('White noise') 

>>> ax_mag.plot(np.arange(-len(sig)+1,len(sig)), autocorr) 

>>> ax_mag.set_title('Autocorrelation') 

>>> fig.tight_layout() 

>>> fig.show() 

 

Gaussian blur implemented using FFT convolution. Notice the dark borders 

around the image, due to the zero-padding beyond its boundaries. 

The `convolve2d` function allows for other types of image boundaries, 

but is far slower. 

 

>>> from scipy import misc 

>>> face = misc.face(gray=True) 

>>> kernel = np.outer(signal.gaussian(70, 8), signal.gaussian(70, 8)) 

>>> blurred = signal.fftconvolve(face, kernel, mode='same') 

 

>>> fig, (ax_orig, ax_kernel, ax_blurred) = plt.subplots(3, 1, 

... figsize=(6, 15)) 

>>> ax_orig.imshow(face, cmap='gray') 

>>> ax_orig.set_title('Original') 

>>> ax_orig.set_axis_off() 

>>> ax_kernel.imshow(kernel, cmap='gray') 

>>> ax_kernel.set_title('Gaussian kernel') 

>>> ax_kernel.set_axis_off() 

>>> ax_blurred.imshow(blurred, cmap='gray') 

>>> ax_blurred.set_title('Blurred') 

>>> ax_blurred.set_axis_off() 

>>> fig.show() 

 

""" 

in1 = asarray(in1) 

in2 = asarray(in2) 

 

if in1.ndim == in2.ndim == 0: # scalar inputs 

return in1 * in2 

elif not in1.ndim == in2.ndim: 

raise ValueError("in1 and in2 should have the same dimensionality") 

elif in1.size == 0 or in2.size == 0: # empty arrays 

return array([]) 

 

s1 = array(in1.shape) 

s2 = array(in2.shape) 

complex_result = (np.issubdtype(in1.dtype, np.complexfloating) or 

np.issubdtype(in2.dtype, np.complexfloating)) 

shape = s1 + s2 - 1 

 

# Check that input sizes are compatible with 'valid' mode 

if _inputs_swap_needed(mode, s1, s2): 

# Convolution is commutative; order doesn't have any effect on output 

in1, s1, in2, s2 = in2, s2, in1, s1 

 

# Speed up FFT by padding to optimal size for FFTPACK 

fshape = [fftpack.helper.next_fast_len(int(d)) for d in shape] 

fslice = tuple([slice(0, int(sz)) for sz in shape]) 

# Pre-1.9 NumPy FFT routines are not threadsafe. For older NumPys, make 

# sure we only call rfftn/irfftn from one thread at a time. 

if not complex_result and (_rfft_mt_safe or _rfft_lock.acquire(False)): 

try: 

sp1 = np.fft.rfftn(in1, fshape) 

sp2 = np.fft.rfftn(in2, fshape) 

ret = (np.fft.irfftn(sp1 * sp2, fshape)[fslice].copy()) 

finally: 

if not _rfft_mt_safe: 

_rfft_lock.release() 

else: 

# If we're here, it's either because we need a complex result, or we 

# failed to acquire _rfft_lock (meaning rfftn isn't threadsafe and 

# is already in use by another thread). In either case, use the 

# (threadsafe but slower) SciPy complex-FFT routines instead. 

sp1 = fftpack.fftn(in1, fshape) 

sp2 = fftpack.fftn(in2, fshape) 

ret = fftpack.ifftn(sp1 * sp2)[fslice].copy() 

if not complex_result: 

ret = ret.real 

 

if mode == "full": 

return ret 

elif mode == "same": 

return _centered(ret, s1) 

elif mode == "valid": 

return _centered(ret, s1 - s2 + 1) 

else: 

raise ValueError("Acceptable mode flags are 'valid'," 

" 'same', or 'full'.") 

 

 

def _numeric_arrays(arrays, kinds='buifc'): 

""" 

See if a list of arrays are all numeric. 

 

Parameters 

---------- 

ndarrays : array or list of arrays 

arrays to check if numeric. 

numeric_kinds : string-like 

The dtypes of the arrays to be checked. If the dtype.kind of 

the ndarrays are not in this string the function returns False and 

otherwise returns True. 

""" 

if type(arrays) == ndarray: 

return arrays.dtype.kind in kinds 

for array_ in arrays: 

if array_.dtype.kind not in kinds: 

return False 

return True 

 

 

def _prod(iterable): 

""" 

Product of a list of numbers. 

Faster than np.prod for short lists like array shapes. 

""" 

product = 1 

for x in iterable: 

product *= x 

return product 

 

 

def _fftconv_faster(x, h, mode): 

""" 

See if using `fftconvolve` or `_correlateND` is faster. The boolean value 

returned depends on the sizes and shapes of the input values. 

 

The big O ratios were found to hold across different machines, which makes 

sense as it's the ratio that matters (the effective speed of the computer 

is found in both big O constants). Regardless, this had been tuned on an 

early 2015 MacBook Pro with 8GB RAM and an Intel i5 processor. 

""" 

if mode == 'full': 

out_shape = [n + k - 1 for n, k in zip(x.shape, h.shape)] 

big_O_constant = 10963.92823819 if x.ndim == 1 else 8899.1104874 

elif mode == 'same': 

out_shape = x.shape 

if x.ndim == 1: 

if h.size <= x.size: 

big_O_constant = 7183.41306773 

else: 

big_O_constant = 856.78174111 

else: 

big_O_constant = 34519.21021589 

elif mode == 'valid': 

out_shape = [n - k + 1 for n, k in zip(x.shape, h.shape)] 

big_O_constant = 41954.28006344 if x.ndim == 1 else 66453.24316434 

else: 

raise ValueError("Acceptable mode flags are 'valid'," 

" 'same', or 'full'.") 

 

# see whether the Fourier transform convolution method or the direct 

# convolution method is faster (discussed in scikit-image PR #1792) 

direct_time = (x.size * h.size * _prod(out_shape)) 

fft_time = sum(n * math.log(n) for n in (x.shape + h.shape + 

tuple(out_shape))) 

return big_O_constant * fft_time < direct_time 

 

 

def _reverse_and_conj(x): 

""" 

Reverse array `x` in all dimensions and perform the complex conjugate 

""" 

reverse = [slice(None, None, -1)] * x.ndim 

return x[reverse].conj() 

 

 

def _np_conv_ok(volume, kernel, mode): 

""" 

See if numpy supports convolution of `volume` and `kernel` (i.e. both are 

1D ndarrays and of the appropriate shape). Numpy's 'same' mode uses the 

size of the larger input, while Scipy's uses the size of the first input. 

 

Invalid mode strings will return False and be caught by the calling func. 

""" 

if volume.ndim == kernel.ndim == 1: 

if mode in ('full', 'valid'): 

return True 

elif mode == 'same': 

return volume.size >= kernel.size 

else: 

return False 

 

 

def _timeit_fast(stmt="pass", setup="pass", repeat=3): 

""" 

Returns the time the statement/function took, in seconds. 

 

Faster, less precise version of IPython's timeit. `stmt` can be a statement 

written as a string or a callable. 

 

Will do only 1 loop (like IPython's timeit) with no repetitions 

(unlike IPython) for very slow functions. For fast functions, only does 

enough loops to take 5 ms, which seems to produce similar results (on 

Windows at least), and avoids doing an extraneous cycle that isn't 

measured. 

 

""" 

timer = timeit.Timer(stmt, setup) 

 

# determine number of calls per rep so total time for 1 rep >= 5 ms 

x = 0 

for p in range(0, 10): 

number = 10**p 

x = timer.timeit(number) # seconds 

if x >= 5e-3 / 10: # 5 ms for final test, 1/10th that for this one 

break 

if x > 1: # second 

# If it's macroscopic, don't bother with repetitions 

best = x 

else: 

number *= 10 

r = timer.repeat(repeat, number) 

best = min(r) 

 

sec = best / number 

return sec 

 

 

def choose_conv_method(in1, in2, mode='full', measure=False): 

""" 

Find the fastest convolution/correlation method. 

 

This primarily exists to be called during the ``method='auto'`` option in 

`convolve` and `correlate`, but can also be used when performing many 

convolutions of the same input shapes and dtypes, determining 

which method to use for all of them, either to avoid the overhead of the 

'auto' option or to use accurate real-world measurements. 

 

Parameters 

---------- 

in1 : array_like 

The first argument passed into the convolution function. 

in2 : array_like 

The second argument passed into the convolution function. 

mode : str {'full', 'valid', 'same'}, optional 

A string indicating the size of the output: 

 

``full`` 

The output is the full discrete linear convolution 

of the inputs. (Default) 

``valid`` 

The output consists only of those elements that do not 

rely on the zero-padding. 

``same`` 

The output is the same size as `in1`, centered 

with respect to the 'full' output. 

measure : bool, optional 

If True, run and time the convolution of `in1` and `in2` with both 

methods and return the fastest. If False (default), predict the fastest 

method using precomputed values. 

 

Returns 

------- 

method : str 

A string indicating which convolution method is fastest, either 

'direct' or 'fft' 

times : dict, optional 

A dictionary containing the times (in seconds) needed for each method. 

This value is only returned if ``measure=True``. 

 

See Also 

-------- 

convolve 

correlate 

 

Notes 

----- 

For large n, ``measure=False`` is accurate and can quickly determine the 

fastest method to perform the convolution. However, this is not as 

accurate for small n (when any dimension in the input or output is small). 

 

In practice, we found that this function estimates the faster method up to 

a multiplicative factor of 5 (i.e., the estimated method is *at most* 5 

times slower than the fastest method). The estimation values were tuned on 

an early 2015 MacBook Pro with 8GB RAM but we found that the prediction 

held *fairly* accurately across different machines. 

 

If ``measure=True``, time the convolutions. Because this function uses 

`fftconvolve`, an error will be thrown if it does not support the inputs. 

There are cases when `fftconvolve` supports the inputs but this function 

returns `direct` (e.g., to protect against floating point integer 

precision). 

 

.. versionadded:: 0.19 

 

Examples 

-------- 

Estimate the fastest method for a given input: 

 

>>> from scipy import signal 

>>> a = np.random.randn(1000) 

>>> b = np.random.randn(1000000) 

>>> method = signal.choose_conv_method(a, b, mode='same') 

>>> method 

'fft' 

 

This can then be applied to other arrays of the same dtype and shape: 

 

>>> c = np.random.randn(1000) 

>>> d = np.random.randn(1000000) 

>>> # `method` works with correlate and convolve 

>>> corr1 = signal.correlate(a, b, mode='same', method=method) 

>>> corr2 = signal.correlate(c, d, mode='same', method=method) 

>>> conv1 = signal.convolve(a, b, mode='same', method=method) 

>>> conv2 = signal.convolve(c, d, mode='same', method=method) 

 

""" 

volume = asarray(in1) 

kernel = asarray(in2) 

 

if measure: 

times = {} 

for method in ['fft', 'direct']: 

times[method] = _timeit_fast(lambda: convolve(volume, kernel, 

mode=mode, method=method)) 

 

chosen_method = 'fft' if times['fft'] < times['direct'] else 'direct' 

return chosen_method, times 

 

# fftconvolve doesn't support complex256 

fftconv_unsup = "complex256" if sys.maxsize > 2**32 else "complex192" 

if hasattr(np, fftconv_unsup): 

if volume.dtype == fftconv_unsup or kernel.dtype == fftconv_unsup: 

return 'direct' 

 

# for integer input, 

# catch when more precision required than float provides (representing an 

# integer as float can lose precision in fftconvolve if larger than 2**52) 

if any([_numeric_arrays([x], kinds='ui') for x in [volume, kernel]]): 

max_value = int(np.abs(volume).max()) * int(np.abs(kernel).max()) 

max_value *= int(min(volume.size, kernel.size)) 

if max_value > 2**np.finfo('float').nmant - 1: 

return 'direct' 

 

if _numeric_arrays([volume, kernel], kinds='b'): 

return 'direct' 

 

if _numeric_arrays([volume, kernel]): 

if _fftconv_faster(volume, kernel, mode): 

return 'fft' 

 

return 'direct' 

 

 

def convolve(in1, in2, mode='full', method='auto'): 

""" 

Convolve two N-dimensional arrays. 

 

Convolve `in1` and `in2`, with the output size determined by the 

`mode` argument. 

 

Parameters 

---------- 

in1 : array_like 

First input. 

in2 : array_like 

Second input. Should have the same number of dimensions as `in1`. 

mode : str {'full', 'valid', 'same'}, optional 

A string indicating the size of the output: 

 

``full`` 

The output is the full discrete linear convolution 

of the inputs. (Default) 

``valid`` 

The output consists only of those elements that do not 

rely on the zero-padding. In 'valid' mode, either `in1` or `in2` 

must be at least as large as the other in every dimension. 

``same`` 

The output is the same size as `in1`, centered 

with respect to the 'full' output. 

method : str {'auto', 'direct', 'fft'}, optional 

A string indicating which method to use to calculate the convolution. 

 

``direct`` 

The convolution is determined directly from sums, the definition of 

convolution. 

``fft`` 

The Fourier Transform is used to perform the convolution by calling 

`fftconvolve`. 

``auto`` 

Automatically chooses direct or Fourier method based on an estimate 

of which is faster (default). See Notes for more detail. 

 

.. versionadded:: 0.19.0 

 

Returns 

------- 

convolve : array 

An N-dimensional array containing a subset of the discrete linear 

convolution of `in1` with `in2`. 

 

See Also 

-------- 

numpy.polymul : performs polynomial multiplication (same operation, but 

also accepts poly1d objects) 

choose_conv_method : chooses the fastest appropriate convolution method 

fftconvolve 

 

Notes 

----- 

By default, `convolve` and `correlate` use ``method='auto'``, which calls 

`choose_conv_method` to choose the fastest method using pre-computed 

values (`choose_conv_method` can also measure real-world timing with a 

keyword argument). Because `fftconvolve` relies on floating point numbers, 

there are certain constraints that may force `method=direct` (more detail 

in `choose_conv_method` docstring). 

 

Examples 

-------- 

Smooth a square pulse using a Hann window: 

 

>>> from scipy import signal 

>>> sig = np.repeat([0., 1., 0.], 100) 

>>> win = signal.hann(50) 

>>> filtered = signal.convolve(sig, win, mode='same') / sum(win) 

 

>>> import matplotlib.pyplot as plt 

>>> fig, (ax_orig, ax_win, ax_filt) = plt.subplots(3, 1, sharex=True) 

>>> ax_orig.plot(sig) 

>>> ax_orig.set_title('Original pulse') 

>>> ax_orig.margins(0, 0.1) 

>>> ax_win.plot(win) 

>>> ax_win.set_title('Filter impulse response') 

>>> ax_win.margins(0, 0.1) 

>>> ax_filt.plot(filtered) 

>>> ax_filt.set_title('Filtered signal') 

>>> ax_filt.margins(0, 0.1) 

>>> fig.tight_layout() 

>>> fig.show() 

 

""" 

volume = asarray(in1) 

kernel = asarray(in2) 

 

if volume.ndim == kernel.ndim == 0: 

return volume * kernel 

elif volume.ndim != kernel.ndim: 

raise ValueError("volume and kernel should have the same " 

"dimensionality") 

 

if _inputs_swap_needed(mode, volume.shape, kernel.shape): 

# Convolution is commutative; order doesn't have any effect on output 

volume, kernel = kernel, volume 

 

if method == 'auto': 

method = choose_conv_method(volume, kernel, mode=mode) 

 

if method == 'fft': 

out = fftconvolve(volume, kernel, mode=mode) 

result_type = np.result_type(volume, kernel) 

if result_type.kind in {'u', 'i'}: 

out = np.around(out) 

return out.astype(result_type) 

elif method == 'direct': 

# fastpath to faster numpy.convolve for 1d inputs when possible 

if _np_conv_ok(volume, kernel, mode): 

return np.convolve(volume, kernel, mode) 

 

return correlate(volume, _reverse_and_conj(kernel), mode, 'direct') 

else: 

raise ValueError("Acceptable method flags are 'auto'," 

" 'direct', or 'fft'.") 

 

 

def order_filter(a, domain, rank): 

""" 

Perform an order filter on an N-dimensional array. 

 

Perform an order filter on the array in. The domain argument acts as a 

mask centered over each pixel. The non-zero elements of domain are 

used to select elements surrounding each input pixel which are placed 

in a list. The list is sorted, and the output for that pixel is the 

element corresponding to rank in the sorted list. 

 

Parameters 

---------- 

a : ndarray 

The N-dimensional input array. 

domain : array_like 

A mask array with the same number of dimensions as `a`. 

Each dimension should have an odd number of elements. 

rank : int 

A non-negative integer which selects the element from the 

sorted list (0 corresponds to the smallest element, 1 is the 

next smallest element, etc.). 

 

Returns 

------- 

out : ndarray 

The results of the order filter in an array with the same 

shape as `a`. 

 

Examples 

-------- 

>>> from scipy import signal 

>>> x = np.arange(25).reshape(5, 5) 

>>> domain = np.identity(3) 

>>> x 

array([[ 0, 1, 2, 3, 4], 

[ 5, 6, 7, 8, 9], 

[10, 11, 12, 13, 14], 

[15, 16, 17, 18, 19], 

[20, 21, 22, 23, 24]]) 

>>> signal.order_filter(x, domain, 0) 

array([[ 0., 0., 0., 0., 0.], 

[ 0., 0., 1., 2., 0.], 

[ 0., 5., 6., 7., 0.], 

[ 0., 10., 11., 12., 0.], 

[ 0., 0., 0., 0., 0.]]) 

>>> signal.order_filter(x, domain, 2) 

array([[ 6., 7., 8., 9., 4.], 

[ 11., 12., 13., 14., 9.], 

[ 16., 17., 18., 19., 14.], 

[ 21., 22., 23., 24., 19.], 

[ 20., 21., 22., 23., 24.]]) 

 

""" 

domain = asarray(domain) 

size = domain.shape 

for k in range(len(size)): 

if (size[k] % 2) != 1: 

raise ValueError("Each dimension of domain argument " 

" should have an odd number of elements.") 

return sigtools._order_filterND(a, domain, rank) 

 

 

def medfilt(volume, kernel_size=None): 

""" 

Perform a median filter on an N-dimensional array. 

 

Apply a median filter to the input array using a local window-size 

given by `kernel_size`. 

 

Parameters 

---------- 

volume : array_like 

An N-dimensional input array. 

kernel_size : array_like, optional 

A scalar or an N-length list giving the size of the median filter 

window in each dimension. Elements of `kernel_size` should be odd. 

If `kernel_size` is a scalar, then this scalar is used as the size in 

each dimension. Default size is 3 for each dimension. 

 

Returns 

------- 

out : ndarray 

An array the same size as input containing the median filtered 

result. 

 

""" 

volume = atleast_1d(volume) 

if kernel_size is None: 

kernel_size = [3] * volume.ndim 

kernel_size = asarray(kernel_size) 

if kernel_size.shape == (): 

kernel_size = np.repeat(kernel_size.item(), volume.ndim) 

 

for k in range(volume.ndim): 

if (kernel_size[k] % 2) != 1: 

raise ValueError("Each element of kernel_size should be odd.") 

 

domain = ones(kernel_size) 

 

numels = product(kernel_size, axis=0) 

order = numels // 2 

return sigtools._order_filterND(volume, domain, order) 

 

 

def wiener(im, mysize=None, noise=None): 

""" 

Perform a Wiener filter on an N-dimensional array. 

 

Apply a Wiener filter to the N-dimensional array `im`. 

 

Parameters 

---------- 

im : ndarray 

An N-dimensional array. 

mysize : int or array_like, optional 

A scalar or an N-length list giving the size of the Wiener filter 

window in each dimension. Elements of mysize should be odd. 

If mysize is a scalar, then this scalar is used as the size 

in each dimension. 

noise : float, optional 

The noise-power to use. If None, then noise is estimated as the 

average of the local variance of the input. 

 

Returns 

------- 

out : ndarray 

Wiener filtered result with the same shape as `im`. 

 

""" 

im = asarray(im) 

if mysize is None: 

mysize = [3] * im.ndim 

mysize = asarray(mysize) 

if mysize.shape == (): 

mysize = np.repeat(mysize.item(), im.ndim) 

 

# Estimate the local mean 

lMean = correlate(im, ones(mysize), 'same') / product(mysize, axis=0) 

 

# Estimate the local variance 

lVar = (correlate(im ** 2, ones(mysize), 'same') / 

product(mysize, axis=0) - lMean ** 2) 

 

# Estimate the noise power if needed. 

if noise is None: 

noise = mean(ravel(lVar), axis=0) 

 

res = (im - lMean) 

res *= (1 - noise / lVar) 

res += lMean 

out = where(lVar < noise, lMean, res) 

 

return out 

 

 

def convolve2d(in1, in2, mode='full', boundary='fill', fillvalue=0): 

""" 

Convolve two 2-dimensional arrays. 

 

Convolve `in1` and `in2` with output size determined by `mode`, and 

boundary conditions determined by `boundary` and `fillvalue`. 

 

Parameters 

---------- 

in1 : array_like 

First input. 

in2 : array_like 

Second input. Should have the same number of dimensions as `in1`. 

mode : str {'full', 'valid', 'same'}, optional 

A string indicating the size of the output: 

 

``full`` 

The output is the full discrete linear convolution 

of the inputs. (Default) 

``valid`` 

The output consists only of those elements that do not 

rely on the zero-padding. In 'valid' mode, either `in1` or `in2` 

must be at least as large as the other in every dimension. 

``same`` 

The output is the same size as `in1`, centered 

with respect to the 'full' output. 

boundary : str {'fill', 'wrap', 'symm'}, optional 

A flag indicating how to handle boundaries: 

 

``fill`` 

pad input arrays with fillvalue. (default) 

``wrap`` 

circular boundary conditions. 

``symm`` 

symmetrical boundary conditions. 

 

fillvalue : scalar, optional 

Value to fill pad input arrays with. Default is 0. 

 

Returns 

------- 

out : ndarray 

A 2-dimensional array containing a subset of the discrete linear 

convolution of `in1` with `in2`. 

 

Examples 

-------- 

Compute the gradient of an image by 2D convolution with a complex Scharr 

operator. (Horizontal operator is real, vertical is imaginary.) Use 

symmetric boundary condition to avoid creating edges at the image 

boundaries. 

 

>>> from scipy import signal 

>>> from scipy import misc 

>>> ascent = misc.ascent() 

>>> scharr = np.array([[ -3-3j, 0-10j, +3 -3j], 

... [-10+0j, 0+ 0j, +10 +0j], 

... [ -3+3j, 0+10j, +3 +3j]]) # Gx + j*Gy 

>>> grad = signal.convolve2d(ascent, scharr, boundary='symm', mode='same') 

 

>>> import matplotlib.pyplot as plt 

>>> fig, (ax_orig, ax_mag, ax_ang) = plt.subplots(3, 1, figsize=(6, 15)) 

>>> ax_orig.imshow(ascent, cmap='gray') 

>>> ax_orig.set_title('Original') 

>>> ax_orig.set_axis_off() 

>>> ax_mag.imshow(np.absolute(grad), cmap='gray') 

>>> ax_mag.set_title('Gradient magnitude') 

>>> ax_mag.set_axis_off() 

>>> ax_ang.imshow(np.angle(grad), cmap='hsv') # hsv is cyclic, like angles 

>>> ax_ang.set_title('Gradient orientation') 

>>> ax_ang.set_axis_off() 

>>> fig.show() 

 

""" 

in1 = asarray(in1) 

in2 = asarray(in2) 

 

if not in1.ndim == in2.ndim == 2: 

raise ValueError('convolve2d inputs must both be 2D arrays') 

 

if _inputs_swap_needed(mode, in1.shape, in2.shape): 

in1, in2 = in2, in1 

 

val = _valfrommode(mode) 

bval = _bvalfromboundary(boundary) 

out = sigtools._convolve2d(in1, in2, 1, val, bval, fillvalue) 

return out 

 

 

def correlate2d(in1, in2, mode='full', boundary='fill', fillvalue=0): 

""" 

Cross-correlate two 2-dimensional arrays. 

 

Cross correlate `in1` and `in2` with output size determined by `mode`, and 

boundary conditions determined by `boundary` and `fillvalue`. 

 

Parameters 

---------- 

in1 : array_like 

First input. 

in2 : array_like 

Second input. Should have the same number of dimensions as `in1`. 

mode : str {'full', 'valid', 'same'}, optional 

A string indicating the size of the output: 

 

``full`` 

The output is the full discrete linear cross-correlation 

of the inputs. (Default) 

``valid`` 

The output consists only of those elements that do not 

rely on the zero-padding. In 'valid' mode, either `in1` or `in2` 

must be at least as large as the other in every dimension. 

``same`` 

The output is the same size as `in1`, centered 

with respect to the 'full' output. 

boundary : str {'fill', 'wrap', 'symm'}, optional 

A flag indicating how to handle boundaries: 

 

``fill`` 

pad input arrays with fillvalue. (default) 

``wrap`` 

circular boundary conditions. 

``symm`` 

symmetrical boundary conditions. 

 

fillvalue : scalar, optional 

Value to fill pad input arrays with. Default is 0. 

 

Returns 

------- 

correlate2d : ndarray 

A 2-dimensional array containing a subset of the discrete linear 

cross-correlation of `in1` with `in2`. 

 

Examples 

-------- 

Use 2D cross-correlation to find the location of a template in a noisy 

image: 

 

>>> from scipy import signal 

>>> from scipy import misc 

>>> face = misc.face(gray=True) - misc.face(gray=True).mean() 

>>> template = np.copy(face[300:365, 670:750]) # right eye 

>>> template -= template.mean() 

>>> face = face + np.random.randn(*face.shape) * 50 # add noise 

>>> corr = signal.correlate2d(face, template, boundary='symm', mode='same') 

>>> y, x = np.unravel_index(np.argmax(corr), corr.shape) # find the match 

 

>>> import matplotlib.pyplot as plt 

>>> fig, (ax_orig, ax_template, ax_corr) = plt.subplots(3, 1, 

... figsize=(6, 15)) 

>>> ax_orig.imshow(face, cmap='gray') 

>>> ax_orig.set_title('Original') 

>>> ax_orig.set_axis_off() 

>>> ax_template.imshow(template, cmap='gray') 

>>> ax_template.set_title('Template') 

>>> ax_template.set_axis_off() 

>>> ax_corr.imshow(corr, cmap='gray') 

>>> ax_corr.set_title('Cross-correlation') 

>>> ax_corr.set_axis_off() 

>>> ax_orig.plot(x, y, 'ro') 

>>> fig.show() 

 

""" 

in1 = asarray(in1) 

in2 = asarray(in2) 

 

if not in1.ndim == in2.ndim == 2: 

raise ValueError('correlate2d inputs must both be 2D arrays') 

 

swapped_inputs = _inputs_swap_needed(mode, in1.shape, in2.shape) 

if swapped_inputs: 

in1, in2 = in2, in1 

 

val = _valfrommode(mode) 

bval = _bvalfromboundary(boundary) 

out = sigtools._convolve2d(in1, in2.conj(), 0, val, bval, fillvalue) 

 

if swapped_inputs: 

out = out[::-1, ::-1] 

 

return out 

 

 

def medfilt2d(input, kernel_size=3): 

""" 

Median filter a 2-dimensional array. 

 

Apply a median filter to the `input` array using a local window-size 

given by `kernel_size` (must be odd). 

 

Parameters 

---------- 

input : array_like 

A 2-dimensional input array. 

kernel_size : array_like, optional 

A scalar or a list of length 2, giving the size of the 

median filter window in each dimension. Elements of 

`kernel_size` should be odd. If `kernel_size` is a scalar, 

then this scalar is used as the size in each dimension. 

Default is a kernel of size (3, 3). 

 

Returns 

------- 

out : ndarray 

An array the same size as input containing the median filtered 

result. 

 

""" 

image = asarray(input) 

if kernel_size is None: 

kernel_size = [3] * 2 

kernel_size = asarray(kernel_size) 

if kernel_size.shape == (): 

kernel_size = np.repeat(kernel_size.item(), 2) 

 

for size in kernel_size: 

if (size % 2) != 1: 

raise ValueError("Each element of kernel_size should be odd.") 

 

return sigtools._medfilt2d(image, kernel_size) 

 

 

def lfilter(b, a, x, axis=-1, zi=None): 

""" 

Filter data along one-dimension with an IIR or FIR filter. 

 

Filter a data sequence, `x`, using a digital filter. This works for many 

fundamental data types (including Object type). The filter is a direct 

form II transposed implementation of the standard difference equation 

(see Notes). 

 

Parameters 

---------- 

b : array_like 

The numerator coefficient vector in a 1-D sequence. 

a : array_like 

The denominator coefficient vector in a 1-D sequence. If ``a[0]`` 

is not 1, then both `a` and `b` are normalized by ``a[0]``. 

x : array_like 

An N-dimensional input array. 

axis : int, optional 

The axis of the input data array along which to apply the 

linear filter. The filter is applied to each subarray along 

this axis. Default is -1. 

zi : array_like, optional 

Initial conditions for the filter delays. It is a vector 

(or array of vectors for an N-dimensional input) of length 

``max(len(a), len(b)) - 1``. If `zi` is None or is not given then 

initial rest is assumed. See `lfiltic` for more information. 

 

Returns 

------- 

y : array 

The output of the digital filter. 

zf : array, optional 

If `zi` is None, this is not returned, otherwise, `zf` holds the 

final filter delay values. 

 

See Also 

-------- 

lfiltic : Construct initial conditions for `lfilter`. 

lfilter_zi : Compute initial state (steady state of step response) for 

`lfilter`. 

filtfilt : A forward-backward filter, to obtain a filter with linear phase. 

savgol_filter : A Savitzky-Golay filter. 

sosfilt: Filter data using cascaded second-order sections. 

sosfiltfilt: A forward-backward filter using second-order sections. 

 

Notes 

----- 

The filter function is implemented as a direct II transposed structure. 

This means that the filter implements:: 

 

a[0]*y[n] = b[0]*x[n] + b[1]*x[n-1] + ... + b[M]*x[n-M] 

- a[1]*y[n-1] - ... - a[N]*y[n-N] 

 

where `M` is the degree of the numerator, `N` is the degree of the 

denominator, and `n` is the sample number. It is implemented using 

the following difference equations (assuming M = N):: 

 

a[0]*y[n] = b[0] * x[n] + d[0][n-1] 

d[0][n] = b[1] * x[n] - a[1] * y[n] + d[1][n-1] 

d[1][n] = b[2] * x[n] - a[2] * y[n] + d[2][n-1] 

... 

d[N-2][n] = b[N-1]*x[n] - a[N-1]*y[n] + d[N-1][n-1] 

d[N-1][n] = b[N] * x[n] - a[N] * y[n] 

 

where `d` are the state variables. 

 

The rational transfer function describing this filter in the 

z-transform domain is:: 

 

-1 -M 

b[0] + b[1]z + ... + b[M] z 

Y(z) = -------------------------------- X(z) 

-1 -N 

a[0] + a[1]z + ... + a[N] z 

 

Examples 

-------- 

Generate a noisy signal to be filtered: 

 

>>> from scipy import signal 

>>> import matplotlib.pyplot as plt 

>>> t = np.linspace(-1, 1, 201) 

>>> x = (np.sin(2*np.pi*0.75*t*(1-t) + 2.1) + 

... 0.1*np.sin(2*np.pi*1.25*t + 1) + 

... 0.18*np.cos(2*np.pi*3.85*t)) 

>>> xn = x + np.random.randn(len(t)) * 0.08 

 

Create an order 3 lowpass butterworth filter: 

 

>>> b, a = signal.butter(3, 0.05) 

 

Apply the filter to xn. Use lfilter_zi to choose the initial condition of 

the filter: 

 

>>> zi = signal.lfilter_zi(b, a) 

>>> z, _ = signal.lfilter(b, a, xn, zi=zi*xn[0]) 

 

Apply the filter again, to have a result filtered at an order the same as 

filtfilt: 

 

>>> z2, _ = signal.lfilter(b, a, z, zi=zi*z[0]) 

 

Use filtfilt to apply the filter: 

 

>>> y = signal.filtfilt(b, a, xn) 

 

Plot the original signal and the various filtered versions: 

 

>>> plt.figure 

>>> plt.plot(t, xn, 'b', alpha=0.75) 

>>> plt.plot(t, z, 'r--', t, z2, 'r', t, y, 'k') 

>>> plt.legend(('noisy signal', 'lfilter, once', 'lfilter, twice', 

... 'filtfilt'), loc='best') 

>>> plt.grid(True) 

>>> plt.show() 

 

""" 

a = np.atleast_1d(a) 

if len(a) == 1: 

# This path only supports types fdgFDGO to mirror _linear_filter below. 

# Any of b, a, x, or zi can set the dtype, but there is no default 

# casting of other types; instead a NotImplementedError is raised. 

b = np.asarray(b) 

a = np.asarray(a) 

if b.ndim != 1 and a.ndim != 1: 

raise ValueError('object of too small depth for desired array') 

x = np.asarray(x) 

inputs = [b, a, x] 

if zi is not None: 

# _linear_filter does not broadcast zi, but does do expansion of 

# singleton dims. 

zi = np.asarray(zi) 

if zi.ndim != x.ndim: 

raise ValueError('object of too small depth for desired array') 

expected_shape = list(x.shape) 

expected_shape[axis] = b.shape[0] - 1 

expected_shape = tuple(expected_shape) 

# check the trivial case where zi is the right shape first 

if zi.shape != expected_shape: 

strides = zi.ndim * [None] 

if axis < 0: 

axis += zi.ndim 

for k in range(zi.ndim): 

if k == axis and zi.shape[k] == expected_shape[k]: 

strides[k] = zi.strides[k] 

elif k != axis and zi.shape[k] == expected_shape[k]: 

strides[k] = zi.strides[k] 

elif k != axis and zi.shape[k] == 1: 

strides[k] = 0 

else: 

raise ValueError('Unexpected shape for zi: expected ' 

'%s, found %s.' % 

(expected_shape, zi.shape)) 

zi = np.lib.stride_tricks.as_strided(zi, expected_shape, 

strides) 

inputs.append(zi) 

dtype = np.result_type(*inputs) 

 

if dtype.char not in 'fdgFDGO': 

raise NotImplementedError("input type '%s' not supported" % dtype) 

 

b = np.array(b, dtype=dtype) 

a = np.array(a, dtype=dtype, copy=False) 

b /= a[0] 

x = np.array(x, dtype=dtype, copy=False) 

 

out_full = np.apply_along_axis(lambda y: np.convolve(b, y), axis, x) 

ind = out_full.ndim * [slice(None)] 

if zi is not None: 

ind[axis] = slice(zi.shape[axis]) 

out_full[ind] += zi 

 

ind[axis] = slice(out_full.shape[axis] - len(b) + 1) 

out = out_full[ind] 

 

if zi is None: 

return out 

else: 

ind[axis] = slice(out_full.shape[axis] - len(b) + 1, None) 

zf = out_full[ind] 

return out, zf 

else: 

if zi is None: 

return sigtools._linear_filter(b, a, x, axis) 

else: 

return sigtools._linear_filter(b, a, x, axis, zi) 

 

 

def lfiltic(b, a, y, x=None): 

""" 

Construct initial conditions for lfilter given input and output vectors. 

 

Given a linear filter (b, a) and initial conditions on the output `y` 

and the input `x`, return the initial conditions on the state vector zi 

which is used by `lfilter` to generate the output given the input. 

 

Parameters 

---------- 

b : array_like 

Linear filter term. 

a : array_like 

Linear filter term. 

y : array_like 

Initial conditions. 

 

If ``N = len(a) - 1``, then ``y = {y[-1], y[-2], ..., y[-N]}``. 

 

If `y` is too short, it is padded with zeros. 

x : array_like, optional 

Initial conditions. 

 

If ``M = len(b) - 1``, then ``x = {x[-1], x[-2], ..., x[-M]}``. 

 

If `x` is not given, its initial conditions are assumed zero. 

 

If `x` is too short, it is padded with zeros. 

 

Returns 

------- 

zi : ndarray 

The state vector ``zi = {z_0[-1], z_1[-1], ..., z_K-1[-1]}``, 

where ``K = max(M, N)``. 

 

See Also 

-------- 

lfilter, lfilter_zi 

 

""" 

N = np.size(a) - 1 

M = np.size(b) - 1 

K = max(M, N) 

y = asarray(y) 

if y.dtype.kind in 'bui': 

# ensure calculations are floating point 

y = y.astype(np.float64) 

zi = zeros(K, y.dtype) 

if x is None: 

x = zeros(M, y.dtype) 

else: 

x = asarray(x) 

L = np.size(x) 

if L < M: 

x = r_[x, zeros(M - L)] 

L = np.size(y) 

if L < N: 

y = r_[y, zeros(N - L)] 

 

for m in range(M): 

zi[m] = np.sum(b[m + 1:] * x[:M - m], axis=0) 

 

for m in range(N): 

zi[m] -= np.sum(a[m + 1:] * y[:N - m], axis=0) 

 

return zi 

 

 

def deconvolve(signal, divisor): 

"""Deconvolves ``divisor`` out of ``signal`` using inverse filtering. 

 

Returns the quotient and remainder such that 

``signal = convolve(divisor, quotient) + remainder`` 

 

Parameters 

---------- 

signal : array_like 

Signal data, typically a recorded signal 

divisor : array_like 

Divisor data, typically an impulse response or filter that was 

applied to the original signal 

 

Returns 

------- 

quotient : ndarray 

Quotient, typically the recovered original signal 

remainder : ndarray 

Remainder 

 

Examples 

-------- 

Deconvolve a signal that's been filtered: 

 

>>> from scipy import signal 

>>> original = [0, 1, 0, 0, 1, 1, 0, 0] 

>>> impulse_response = [2, 1] 

>>> recorded = signal.convolve(impulse_response, original) 

>>> recorded 

array([0, 2, 1, 0, 2, 3, 1, 0, 0]) 

>>> recovered, remainder = signal.deconvolve(recorded, impulse_response) 

>>> recovered 

array([ 0., 1., 0., 0., 1., 1., 0., 0.]) 

 

See Also 

-------- 

numpy.polydiv : performs polynomial division (same operation, but 

also accepts poly1d objects) 

 

""" 

num = atleast_1d(signal) 

den = atleast_1d(divisor) 

N = len(num) 

D = len(den) 

if D > N: 

quot = [] 

rem = num 

else: 

input = zeros(N - D + 1, float) 

input[0] = 1 

quot = lfilter(num, den, input) 

rem = num - convolve(den, quot, mode='full') 

return quot, rem 

 

 

def hilbert(x, N=None, axis=-1): 

""" 

Compute the analytic signal, using the Hilbert transform. 

 

The transformation is done along the last axis by default. 

 

Parameters 

---------- 

x : array_like 

Signal data. Must be real. 

N : int, optional 

Number of Fourier components. Default: ``x.shape[axis]`` 

axis : int, optional 

Axis along which to do the transformation. Default: -1. 

 

Returns 

------- 

xa : ndarray 

Analytic signal of `x`, of each 1-D array along `axis` 

 

See Also 

-------- 

scipy.fftpack.hilbert : Return Hilbert transform of a periodic sequence x. 

 

Notes 

----- 

The analytic signal ``x_a(t)`` of signal ``x(t)`` is: 

 

.. math:: x_a = F^{-1}(F(x) 2U) = x + i y 

 

where `F` is the Fourier transform, `U` the unit step function, 

and `y` the Hilbert transform of `x`. [1]_ 

 

In other words, the negative half of the frequency spectrum is zeroed 

out, turning the real-valued signal into a complex signal. The Hilbert 

transformed signal can be obtained from ``np.imag(hilbert(x))``, and the 

original signal from ``np.real(hilbert(x))``. 

 

Examples 

--------- 

In this example we use the Hilbert transform to determine the amplitude 

envelope and instantaneous frequency of an amplitude-modulated signal. 

 

>>> import numpy as np 

>>> import matplotlib.pyplot as plt 

>>> from scipy.signal import hilbert, chirp 

 

>>> duration = 1.0 

>>> fs = 400.0 

>>> samples = int(fs*duration) 

>>> t = np.arange(samples) / fs 

 

We create a chirp of which the frequency increases from 20 Hz to 100 Hz and 

apply an amplitude modulation. 

 

>>> signal = chirp(t, 20.0, t[-1], 100.0) 

>>> signal *= (1.0 + 0.5 * np.sin(2.0*np.pi*3.0*t) ) 

 

The amplitude envelope is given by magnitude of the analytic signal. The 

instantaneous frequency can be obtained by differentiating the 

instantaneous phase in respect to time. The instantaneous phase corresponds 

to the phase angle of the analytic signal. 

 

>>> analytic_signal = hilbert(signal) 

>>> amplitude_envelope = np.abs(analytic_signal) 

>>> instantaneous_phase = np.unwrap(np.angle(analytic_signal)) 

>>> instantaneous_frequency = (np.diff(instantaneous_phase) / 

... (2.0*np.pi) * fs) 

 

>>> fig = plt.figure() 

>>> ax0 = fig.add_subplot(211) 

>>> ax0.plot(t, signal, label='signal') 

>>> ax0.plot(t, amplitude_envelope, label='envelope') 

>>> ax0.set_xlabel("time in seconds") 

>>> ax0.legend() 

>>> ax1 = fig.add_subplot(212) 

>>> ax1.plot(t[1:], instantaneous_frequency) 

>>> ax1.set_xlabel("time in seconds") 

>>> ax1.set_ylim(0.0, 120.0) 

 

References 

---------- 

.. [1] Wikipedia, "Analytic signal". 

http://en.wikipedia.org/wiki/Analytic_signal 

.. [2] Leon Cohen, "Time-Frequency Analysis", 1995. Chapter 2. 

.. [3] Alan V. Oppenheim, Ronald W. Schafer. Discrete-Time Signal 

Processing, Third Edition, 2009. Chapter 12. 

ISBN 13: 978-1292-02572-8 

 

""" 

x = asarray(x) 

if iscomplexobj(x): 

raise ValueError("x must be real.") 

if N is None: 

N = x.shape[axis] 

if N <= 0: 

raise ValueError("N must be positive.") 

 

Xf = fftpack.fft(x, N, axis=axis) 

h = zeros(N) 

if N % 2 == 0: 

h[0] = h[N // 2] = 1 

h[1:N // 2] = 2 

else: 

h[0] = 1 

h[1:(N + 1) // 2] = 2 

 

if x.ndim > 1: 

ind = [newaxis] * x.ndim 

ind[axis] = slice(None) 

h = h[ind] 

x = fftpack.ifft(Xf * h, axis=axis) 

return x 

 

 

def hilbert2(x, N=None): 

""" 

Compute the '2-D' analytic signal of `x` 

 

Parameters 

---------- 

x : array_like 

2-D signal data. 

N : int or tuple of two ints, optional 

Number of Fourier components. Default is ``x.shape`` 

 

Returns 

------- 

xa : ndarray 

Analytic signal of `x` taken along axes (0,1). 

 

References 

---------- 

.. [1] Wikipedia, "Analytic signal", 

http://en.wikipedia.org/wiki/Analytic_signal 

 

""" 

x = atleast_2d(x) 

if x.ndim > 2: 

raise ValueError("x must be 2-D.") 

if iscomplexobj(x): 

raise ValueError("x must be real.") 

if N is None: 

N = x.shape 

elif isinstance(N, int): 

if N <= 0: 

raise ValueError("N must be positive.") 

N = (N, N) 

elif len(N) != 2 or np.any(np.asarray(N) <= 0): 

raise ValueError("When given as a tuple, N must hold exactly " 

"two positive integers") 

 

Xf = fftpack.fft2(x, N, axes=(0, 1)) 

h1 = zeros(N[0], 'd') 

h2 = zeros(N[1], 'd') 

for p in range(2): 

h = eval("h%d" % (p + 1)) 

N1 = N[p] 

if N1 % 2 == 0: 

h[0] = h[N1 // 2] = 1 

h[1:N1 // 2] = 2 

else: 

h[0] = 1 

h[1:(N1 + 1) // 2] = 2 

exec("h%d = h" % (p + 1), globals(), locals()) 

 

h = h1[:, newaxis] * h2[newaxis, :] 

k = x.ndim 

while k > 2: 

h = h[:, newaxis] 

k -= 1 

x = fftpack.ifft2(Xf * h, axes=(0, 1)) 

return x 

 

 

def cmplx_sort(p): 

"""Sort roots based on magnitude. 

 

Parameters 

---------- 

p : array_like 

The roots to sort, as a 1-D array. 

 

Returns 

------- 

p_sorted : ndarray 

Sorted roots. 

indx : ndarray 

Array of indices needed to sort the input `p`. 

 

Examples 

-------- 

>>> from scipy import signal 

>>> vals = [1, 4, 1+1.j, 3] 

>>> p_sorted, indx = signal.cmplx_sort(vals) 

>>> p_sorted 

array([1.+0.j, 1.+1.j, 3.+0.j, 4.+0.j]) 

>>> indx 

array([0, 2, 3, 1]) 

 

""" 

p = asarray(p) 

if iscomplexobj(p): 

indx = argsort(abs(p)) 

else: 

indx = argsort(p) 

return take(p, indx, 0), indx 

 

 

def unique_roots(p, tol=1e-3, rtype='min'): 

""" 

Determine unique roots and their multiplicities from a list of roots. 

 

Parameters 

---------- 

p : array_like 

The list of roots. 

tol : float, optional 

The tolerance for two roots to be considered equal. Default is 1e-3. 

rtype : {'max', 'min, 'avg'}, optional 

How to determine the returned root if multiple roots are within 

`tol` of each other. 

 

- 'max': pick the maximum of those roots. 

- 'min': pick the minimum of those roots. 

- 'avg': take the average of those roots. 

 

Returns 

------- 

pout : ndarray 

The list of unique roots, sorted from low to high. 

mult : ndarray 

The multiplicity of each root. 

 

Notes 

----- 

This utility function is not specific to roots but can be used for any 

sequence of values for which uniqueness and multiplicity has to be 

determined. For a more general routine, see `numpy.unique`. 

 

Examples 

-------- 

>>> from scipy import signal 

>>> vals = [0, 1.3, 1.31, 2.8, 1.25, 2.2, 10.3] 

>>> uniq, mult = signal.unique_roots(vals, tol=2e-2, rtype='avg') 

 

Check which roots have multiplicity larger than 1: 

 

>>> uniq[mult > 1] 

array([ 1.305]) 

 

""" 

if rtype in ['max', 'maximum']: 

comproot = np.max 

elif rtype in ['min', 'minimum']: 

comproot = np.min 

elif rtype in ['avg', 'mean']: 

comproot = np.mean 

else: 

raise ValueError("`rtype` must be one of " 

"{'max', 'maximum', 'min', 'minimum', 'avg', 'mean'}") 

p = asarray(p) * 1.0 

tol = abs(tol) 

p, indx = cmplx_sort(p) 

pout = [] 

mult = [] 

indx = -1 

curp = p[0] + 5 * tol 

sameroots = [] 

for k in range(len(p)): 

tr = p[k] 

if abs(tr - curp) < tol: 

sameroots.append(tr) 

curp = comproot(sameroots) 

pout[indx] = curp 

mult[indx] += 1 

else: 

pout.append(tr) 

curp = tr 

sameroots = [tr] 

indx += 1 

mult.append(1) 

return array(pout), array(mult) 

 

 

def invres(r, p, k, tol=1e-3, rtype='avg'): 

""" 

Compute b(s) and a(s) from partial fraction expansion. 

 

If `M` is the degree of numerator `b` and `N` the degree of denominator 

`a`:: 

 

b(s) b[0] s**(M) + b[1] s**(M-1) + ... + b[M] 

H(s) = ------ = ------------------------------------------ 

a(s) a[0] s**(N) + a[1] s**(N-1) + ... + a[N] 

 

then the partial-fraction expansion H(s) is defined as:: 

 

r[0] r[1] r[-1] 

= -------- + -------- + ... + --------- + k(s) 

(s-p[0]) (s-p[1]) (s-p[-1]) 

 

If there are any repeated roots (closer together than `tol`), then H(s) 

has terms like:: 

 

r[i] r[i+1] r[i+n-1] 

-------- + ----------- + ... + ----------- 

(s-p[i]) (s-p[i])**2 (s-p[i])**n 

 

This function is used for polynomials in positive powers of s or z, 

such as analog filters or digital filters in controls engineering. For 

negative powers of z (typical for digital filters in DSP), use `invresz`. 

 

Parameters 

---------- 

r : array_like 

Residues. 

p : array_like 

Poles. 

k : array_like 

Coefficients of the direct polynomial term. 

tol : float, optional 

The tolerance for two roots to be considered equal. Default is 1e-3. 

rtype : {'max', 'min, 'avg'}, optional 

How to determine the returned root if multiple roots are within 

`tol` of each other. 

 

- 'max': pick the maximum of those roots. 

- 'min': pick the minimum of those roots. 

- 'avg': take the average of those roots. 

 

Returns 

------- 

b : ndarray 

Numerator polynomial coefficients. 

a : ndarray 

Denominator polynomial coefficients. 

 

See Also 

-------- 

residue, invresz, unique_roots 

 

""" 

extra = k 

p, indx = cmplx_sort(p) 

r = take(r, indx, 0) 

pout, mult = unique_roots(p, tol=tol, rtype=rtype) 

p = [] 

for k in range(len(pout)): 

p.extend([pout[k]] * mult[k]) 

a = atleast_1d(poly(p)) 

if len(extra) > 0: 

b = polymul(extra, a) 

else: 

b = [0] 

indx = 0 

for k in range(len(pout)): 

temp = [] 

for l in range(len(pout)): 

if l != k: 

temp.extend([pout[l]] * mult[l]) 

for m in range(mult[k]): 

t2 = temp[:] 

t2.extend([pout[k]] * (mult[k] - m - 1)) 

b = polyadd(b, r[indx] * atleast_1d(poly(t2))) 

indx += 1 

b = real_if_close(b) 

while allclose(b[0], 0, rtol=1e-14) and (b.shape[-1] > 1): 

b = b[1:] 

return b, a 

 

 

def residue(b, a, tol=1e-3, rtype='avg'): 

""" 

Compute partial-fraction expansion of b(s) / a(s). 

 

If `M` is the degree of numerator `b` and `N` the degree of denominator 

`a`:: 

 

b(s) b[0] s**(M) + b[1] s**(M-1) + ... + b[M] 

H(s) = ------ = ------------------------------------------ 

a(s) a[0] s**(N) + a[1] s**(N-1) + ... + a[N] 

 

then the partial-fraction expansion H(s) is defined as:: 

 

r[0] r[1] r[-1] 

= -------- + -------- + ... + --------- + k(s) 

(s-p[0]) (s-p[1]) (s-p[-1]) 

 

If there are any repeated roots (closer together than `tol`), then H(s) 

has terms like:: 

 

r[i] r[i+1] r[i+n-1] 

-------- + ----------- + ... + ----------- 

(s-p[i]) (s-p[i])**2 (s-p[i])**n 

 

This function is used for polynomials in positive powers of s or z, 

such as analog filters or digital filters in controls engineering. For 

negative powers of z (typical for digital filters in DSP), use `residuez`. 

 

Parameters 

---------- 

b : array_like 

Numerator polynomial coefficients. 

a : array_like 

Denominator polynomial coefficients. 

 

Returns 

------- 

r : ndarray 

Residues. 

p : ndarray 

Poles. 

k : ndarray 

Coefficients of the direct polynomial term. 

 

See Also 

-------- 

invres, residuez, numpy.poly, unique_roots 

 

""" 

 

b, a = map(asarray, (b, a)) 

rscale = a[0] 

k, b = polydiv(b, a) 

p = roots(a) 

r = p * 0.0 

pout, mult = unique_roots(p, tol=tol, rtype=rtype) 

p = [] 

for n in range(len(pout)): 

p.extend([pout[n]] * mult[n]) 

p = asarray(p) 

# Compute the residue from the general formula 

indx = 0 

for n in range(len(pout)): 

bn = b.copy() 

pn = [] 

for l in range(len(pout)): 

if l != n: 

pn.extend([pout[l]] * mult[l]) 

an = atleast_1d(poly(pn)) 

# bn(s) / an(s) is (s-po[n])**Nn * b(s) / a(s) where Nn is 

# multiplicity of pole at po[n] 

sig = mult[n] 

for m in range(sig, 0, -1): 

if sig > m: 

# compute next derivative of bn(s) / an(s) 

term1 = polymul(polyder(bn, 1), an) 

term2 = polymul(bn, polyder(an, 1)) 

bn = polysub(term1, term2) 

an = polymul(an, an) 

r[indx + m - 1] = (polyval(bn, pout[n]) / polyval(an, pout[n]) / 

factorial(sig - m)) 

indx += sig 

return r / rscale, p, k 

 

 

def residuez(b, a, tol=1e-3, rtype='avg'): 

""" 

Compute partial-fraction expansion of b(z) / a(z). 

 

If `M` is the degree of numerator `b` and `N` the degree of denominator 

`a`:: 

 

b(z) b[0] + b[1] z**(-1) + ... + b[M] z**(-M) 

H(z) = ------ = ------------------------------------------ 

a(z) a[0] + a[1] z**(-1) + ... + a[N] z**(-N) 

 

then the partial-fraction expansion H(z) is defined as:: 

 

r[0] r[-1] 

= --------------- + ... + ---------------- + k[0] + k[1]z**(-1) ... 

(1-p[0]z**(-1)) (1-p[-1]z**(-1)) 

 

If there are any repeated roots (closer than `tol`), then the partial 

fraction expansion has terms like:: 

 

r[i] r[i+1] r[i+n-1] 

-------------- + ------------------ + ... + ------------------ 

(1-p[i]z**(-1)) (1-p[i]z**(-1))**2 (1-p[i]z**(-1))**n 

 

This function is used for polynomials in negative powers of z, 

such as digital filters in DSP. For positive powers, use `residue`. 

 

Parameters 

---------- 

b : array_like 

Numerator polynomial coefficients. 

a : array_like 

Denominator polynomial coefficients. 

 

Returns 

------- 

r : ndarray 

Residues. 

p : ndarray 

Poles. 

k : ndarray 

Coefficients of the direct polynomial term. 

 

See Also 

-------- 

invresz, residue, unique_roots 

 

""" 

b, a = map(asarray, (b, a)) 

gain = a[0] 

brev, arev = b[::-1], a[::-1] 

krev, brev = polydiv(brev, arev) 

if krev == []: 

k = [] 

else: 

k = krev[::-1] 

b = brev[::-1] 

p = roots(a) 

r = p * 0.0 

pout, mult = unique_roots(p, tol=tol, rtype=rtype) 

p = [] 

for n in range(len(pout)): 

p.extend([pout[n]] * mult[n]) 

p = asarray(p) 

# Compute the residue from the general formula (for discrete-time) 

# the polynomial is in z**(-1) and the multiplication is by terms 

# like this (1-p[i] z**(-1))**mult[i]. After differentiation, 

# we must divide by (-p[i])**(m-k) as well as (m-k)! 

indx = 0 

for n in range(len(pout)): 

bn = brev.copy() 

pn = [] 

for l in range(len(pout)): 

if l != n: 

pn.extend([pout[l]] * mult[l]) 

an = atleast_1d(poly(pn))[::-1] 

# bn(z) / an(z) is (1-po[n] z**(-1))**Nn * b(z) / a(z) where Nn is 

# multiplicity of pole at po[n] and b(z) and a(z) are polynomials. 

sig = mult[n] 

for m in range(sig, 0, -1): 

if sig > m: 

# compute next derivative of bn(s) / an(s) 

term1 = polymul(polyder(bn, 1), an) 

term2 = polymul(bn, polyder(an, 1)) 

bn = polysub(term1, term2) 

an = polymul(an, an) 

r[indx + m - 1] = (polyval(bn, 1.0 / pout[n]) / 

polyval(an, 1.0 / pout[n]) / 

factorial(sig - m) / (-pout[n]) ** (sig - m)) 

indx += sig 

return r / gain, p, k 

 

 

def invresz(r, p, k, tol=1e-3, rtype='avg'): 

""" 

Compute b(z) and a(z) from partial fraction expansion. 

 

If `M` is the degree of numerator `b` and `N` the degree of denominator 

`a`:: 

 

b(z) b[0] + b[1] z**(-1) + ... + b[M] z**(-M) 

H(z) = ------ = ------------------------------------------ 

a(z) a[0] + a[1] z**(-1) + ... + a[N] z**(-N) 

 

then the partial-fraction expansion H(z) is defined as:: 

 

r[0] r[-1] 

= --------------- + ... + ---------------- + k[0] + k[1]z**(-1) ... 

(1-p[0]z**(-1)) (1-p[-1]z**(-1)) 

 

If there are any repeated roots (closer than `tol`), then the partial 

fraction expansion has terms like:: 

 

r[i] r[i+1] r[i+n-1] 

-------------- + ------------------ + ... + ------------------ 

(1-p[i]z**(-1)) (1-p[i]z**(-1))**2 (1-p[i]z**(-1))**n 

 

This function is used for polynomials in negative powers of z, 

such as digital filters in DSP. For positive powers, use `invres`. 

 

Parameters 

---------- 

r : array_like 

Residues. 

p : array_like 

Poles. 

k : array_like 

Coefficients of the direct polynomial term. 

tol : float, optional 

The tolerance for two roots to be considered equal. Default is 1e-3. 

rtype : {'max', 'min, 'avg'}, optional 

How to determine the returned root if multiple roots are within 

`tol` of each other. 

 

- 'max': pick the maximum of those roots. 

- 'min': pick the minimum of those roots. 

- 'avg': take the average of those roots. 

 

Returns 

------- 

b : ndarray 

Numerator polynomial coefficients. 

a : ndarray 

Denominator polynomial coefficients. 

 

See Also 

-------- 

residuez, unique_roots, invres 

 

""" 

extra = asarray(k) 

p, indx = cmplx_sort(p) 

r = take(r, indx, 0) 

pout, mult = unique_roots(p, tol=tol, rtype=rtype) 

p = [] 

for k in range(len(pout)): 

p.extend([pout[k]] * mult[k]) 

a = atleast_1d(poly(p)) 

if len(extra) > 0: 

b = polymul(extra, a) 

else: 

b = [0] 

indx = 0 

brev = asarray(b)[::-1] 

for k in range(len(pout)): 

temp = [] 

# Construct polynomial which does not include any of this root 

for l in range(len(pout)): 

if l != k: 

temp.extend([pout[l]] * mult[l]) 

for m in range(mult[k]): 

t2 = temp[:] 

t2.extend([pout[k]] * (mult[k] - m - 1)) 

brev = polyadd(brev, (r[indx] * atleast_1d(poly(t2)))[::-1]) 

indx += 1 

b = real_if_close(brev[::-1]) 

return b, a 

 

 

def resample(x, num, t=None, axis=0, window=None): 

""" 

Resample `x` to `num` samples using Fourier method along the given axis. 

 

The resampled signal starts at the same value as `x` but is sampled 

with a spacing of ``len(x) / num * (spacing of x)``. Because a 

Fourier method is used, the signal is assumed to be periodic. 

 

Parameters 

---------- 

x : array_like 

The data to be resampled. 

num : int 

The number of samples in the resampled signal. 

t : array_like, optional 

If `t` is given, it is assumed to be the sample positions 

associated with the signal data in `x`. 

axis : int, optional 

The axis of `x` that is resampled. Default is 0. 

window : array_like, callable, string, float, or tuple, optional 

Specifies the window applied to the signal in the Fourier 

domain. See below for details. 

 

Returns 

------- 

resampled_x or (resampled_x, resampled_t) 

Either the resampled array, or, if `t` was given, a tuple 

containing the resampled array and the corresponding resampled 

positions. 

 

See Also 

-------- 

decimate : Downsample the signal after applying an FIR or IIR filter. 

resample_poly : Resample using polyphase filtering and an FIR filter. 

 

Notes 

----- 

The argument `window` controls a Fourier-domain window that tapers 

the Fourier spectrum before zero-padding to alleviate ringing in 

the resampled values for sampled signals you didn't intend to be 

interpreted as band-limited. 

 

If `window` is a function, then it is called with a vector of inputs 

indicating the frequency bins (i.e. fftfreq(x.shape[axis]) ). 

 

If `window` is an array of the same length as `x.shape[axis]` it is 

assumed to be the window to be applied directly in the Fourier 

domain (with dc and low-frequency first). 

 

For any other type of `window`, the function `scipy.signal.get_window` 

is called to generate the window. 

 

The first sample of the returned vector is the same as the first 

sample of the input vector. The spacing between samples is changed 

from ``dx`` to ``dx * len(x) / num``. 

 

If `t` is not None, then it represents the old sample positions, 

and the new sample positions will be returned as well as the new 

samples. 

 

As noted, `resample` uses FFT transformations, which can be very 

slow if the number of input or output samples is large and prime; 

see `scipy.fftpack.fft`. 

 

Examples 

-------- 

Note that the end of the resampled data rises to meet the first 

sample of the next cycle: 

 

>>> from scipy import signal 

 

>>> x = np.linspace(0, 10, 20, endpoint=False) 

>>> y = np.cos(-x**2/6.0) 

>>> f = signal.resample(y, 100) 

>>> xnew = np.linspace(0, 10, 100, endpoint=False) 

 

>>> import matplotlib.pyplot as plt 

>>> plt.plot(x, y, 'go-', xnew, f, '.-', 10, y[0], 'ro') 

>>> plt.legend(['data', 'resampled'], loc='best') 

>>> plt.show() 

""" 

x = asarray(x) 

X = fftpack.fft(x, axis=axis) 

Nx = x.shape[axis] 

if window is not None: 

if callable(window): 

W = window(fftpack.fftfreq(Nx)) 

elif isinstance(window, ndarray): 

if window.shape != (Nx,): 

raise ValueError('window must have the same length as data') 

W = window 

else: 

W = fftpack.ifftshift(get_window(window, Nx)) 

newshape = [1] * x.ndim 

newshape[axis] = len(W) 

W.shape = newshape 

X = X * W 

W.shape = (Nx,) 

sl = [slice(None)] * x.ndim 

newshape = list(x.shape) 

newshape[axis] = num 

N = int(np.minimum(num, Nx)) 

Y = zeros(newshape, 'D') 

sl[axis] = slice(0, (N + 1) // 2) 

Y[sl] = X[sl] 

sl[axis] = slice(-(N - 1) // 2, None) 

Y[sl] = X[sl] 

 

if N % 2 == 0: # special treatment if low number of points is even. So far we have set Y[-N/2]=X[-N/2] 

if N < Nx: # if downsampling 

sl[axis] = slice(N//2,N//2+1,None) # select the component at frequency N/2 

Y[sl] += X[sl] # add the component of X at N/2 

elif N < num: # if upsampling 

sl[axis] = slice(num-N//2,num-N//2+1,None) # select the component at frequency -N/2 

Y[sl] /= 2 # halve the component at -N/2 

temp = Y[sl] 

sl[axis] = slice(N//2,N//2+1,None) # select the component at +N/2 

Y[sl] = temp # set that equal to the component at -N/2 

 

y = fftpack.ifft(Y, axis=axis) * (float(num) / float(Nx)) 

 

if x.dtype.char not in ['F', 'D']: 

y = y.real 

 

if t is None: 

return y 

else: 

new_t = arange(0, num) * (t[1] - t[0]) * Nx / float(num) + t[0] 

return y, new_t 

 

 

def resample_poly(x, up, down, axis=0, window=('kaiser', 5.0)): 

""" 

Resample `x` along the given axis using polyphase filtering. 

 

The signal `x` is upsampled by the factor `up`, a zero-phase low-pass 

FIR filter is applied, and then it is downsampled by the factor `down`. 

The resulting sample rate is ``up / down`` times the original sample 

rate. Values beyond the boundary of the signal are assumed to be zero 

during the filtering step. 

 

Parameters 

---------- 

x : array_like 

The data to be resampled. 

up : int 

The upsampling factor. 

down : int 

The downsampling factor. 

axis : int, optional 

The axis of `x` that is resampled. Default is 0. 

window : string, tuple, or array_like, optional 

Desired window to use to design the low-pass filter, or the FIR filter 

coefficients to employ. See below for details. 

 

Returns 

------- 

resampled_x : array 

The resampled array. 

 

See Also 

-------- 

decimate : Downsample the signal after applying an FIR or IIR filter. 

resample : Resample up or down using the FFT method. 

 

Notes 

----- 

This polyphase method will likely be faster than the Fourier method 

in `scipy.signal.resample` when the number of samples is large and 

prime, or when the number of samples is large and `up` and `down` 

share a large greatest common denominator. The length of the FIR 

filter used will depend on ``max(up, down) // gcd(up, down)``, and 

the number of operations during polyphase filtering will depend on 

the filter length and `down` (see `scipy.signal.upfirdn` for details). 

 

The argument `window` specifies the FIR low-pass filter design. 

 

If `window` is an array_like it is assumed to be the FIR filter 

coefficients. Note that the FIR filter is applied after the upsampling 

step, so it should be designed to operate on a signal at a sampling 

frequency higher than the original by a factor of `up//gcd(up, down)`. 

This function's output will be centered with respect to this array, so it 

is best to pass a symmetric filter with an odd number of samples if, as 

is usually the case, a zero-phase filter is desired. 

 

For any other type of `window`, the functions `scipy.signal.get_window` 

and `scipy.signal.firwin` are called to generate the appropriate filter 

coefficients. 

 

The first sample of the returned vector is the same as the first 

sample of the input vector. The spacing between samples is changed 

from ``dx`` to ``dx * down / float(up)``. 

 

Examples 

-------- 

Note that the end of the resampled data rises to meet the first 

sample of the next cycle for the FFT method, and gets closer to zero 

for the polyphase method: 

 

>>> from scipy import signal 

 

>>> x = np.linspace(0, 10, 20, endpoint=False) 

>>> y = np.cos(-x**2/6.0) 

>>> f_fft = signal.resample(y, 100) 

>>> f_poly = signal.resample_poly(y, 100, 20) 

>>> xnew = np.linspace(0, 10, 100, endpoint=False) 

 

>>> import matplotlib.pyplot as plt 

>>> plt.plot(xnew, f_fft, 'b.-', xnew, f_poly, 'r.-') 

>>> plt.plot(x, y, 'ko-') 

>>> plt.plot(10, y[0], 'bo', 10, 0., 'ro') # boundaries 

>>> plt.legend(['resample', 'resamp_poly', 'data'], loc='best') 

>>> plt.show() 

""" 

x = asarray(x) 

if up != int(up): 

raise ValueError("up must be an integer") 

if down != int(down): 

raise ValueError("down must be an integer") 

up = int(up) 

down = int(down) 

if up < 1 or down < 1: 

raise ValueError('up and down must be >= 1') 

 

# Determine our up and down factors 

# Use a rational approimation to save computation time on really long 

# signals 

g_ = gcd(up, down) 

up //= g_ 

down //= g_ 

if up == down == 1: 

return x.copy() 

n_out = x.shape[axis] * up 

n_out = n_out // down + bool(n_out % down) 

 

if isinstance(window, (list, np.ndarray)): 

window = array(window) # use array to force a copy (we modify it) 

if window.ndim > 1: 

raise ValueError('window must be 1-D') 

half_len = (window.size - 1) // 2 

h = window 

else: 

# Design a linear-phase low-pass FIR filter 

max_rate = max(up, down) 

f_c = 1. / max_rate # cutoff of FIR filter (rel. to Nyquist) 

half_len = 10 * max_rate # reasonable cutoff for our sinc-like function 

h = firwin(2 * half_len + 1, f_c, window=window) 

h *= up 

 

# Zero-pad our filter to put the output samples at the center 

n_pre_pad = (down - half_len % down) 

n_post_pad = 0 

n_pre_remove = (half_len + n_pre_pad) // down 

# We should rarely need to do this given our filter lengths... 

while _output_len(len(h) + n_pre_pad + n_post_pad, x.shape[axis], 

up, down) < n_out + n_pre_remove: 

n_post_pad += 1 

h = np.concatenate((np.zeros(n_pre_pad), h, np.zeros(n_post_pad))) 

n_pre_remove_end = n_pre_remove + n_out 

 

# filter then remove excess 

y = upfirdn(h, x, up, down, axis=axis) 

keep = [slice(None), ]*x.ndim 

keep[axis] = slice(n_pre_remove, n_pre_remove_end) 

return y[keep] 

 

 

def vectorstrength(events, period): 

''' 

Determine the vector strength of the events corresponding to the given 

period. 

 

The vector strength is a measure of phase synchrony, how well the 

timing of the events is synchronized to a single period of a periodic 

signal. 

 

If multiple periods are used, calculate the vector strength of each. 

This is called the "resonating vector strength". 

 

Parameters 

---------- 

events : 1D array_like 

An array of time points containing the timing of the events. 

period : float or array_like 

The period of the signal that the events should synchronize to. 

The period is in the same units as `events`. It can also be an array 

of periods, in which case the outputs are arrays of the same length. 

 

Returns 

------- 

strength : float or 1D array 

The strength of the synchronization. 1.0 is perfect synchronization 

and 0.0 is no synchronization. If `period` is an array, this is also 

an array with each element containing the vector strength at the 

corresponding period. 

phase : float or array 

The phase that the events are most strongly synchronized to in radians. 

If `period` is an array, this is also an array with each element 

containing the phase for the corresponding period. 

 

References 

---------- 

van Hemmen, JL, Longtin, A, and Vollmayr, AN. Testing resonating vector 

strength: Auditory system, electric fish, and noise. 

Chaos 21, 047508 (2011); 

:doi:`10.1063/1.3670512`. 

van Hemmen, JL. Vector strength after Goldberg, Brown, and von Mises: 

biological and mathematical perspectives. Biol Cybern. 

2013 Aug;107(4):385-96. :doi:`10.1007/s00422-013-0561-7`. 

van Hemmen, JL and Vollmayr, AN. Resonating vector strength: what happens 

when we vary the "probing" frequency while keeping the spike times 

fixed. Biol Cybern. 2013 Aug;107(4):491-94. 

:doi:`10.1007/s00422-013-0560-8`. 

''' 

events = asarray(events) 

period = asarray(period) 

if events.ndim > 1: 

raise ValueError('events cannot have dimensions more than 1') 

if period.ndim > 1: 

raise ValueError('period cannot have dimensions more than 1') 

 

# we need to know later if period was originally a scalar 

scalarperiod = not period.ndim 

 

events = atleast_2d(events) 

period = atleast_2d(period) 

if (period <= 0).any(): 

raise ValueError('periods must be positive') 

 

# this converts the times to vectors 

vectors = exp(dot(2j*pi/period.T, events)) 

 

# the vector strength is just the magnitude of the mean of the vectors 

# the vector phase is the angle of the mean of the vectors 

vectormean = mean(vectors, axis=1) 

strength = abs(vectormean) 

phase = angle(vectormean) 

 

# if the original period was a scalar, return scalars 

if scalarperiod: 

strength = strength[0] 

phase = phase[0] 

return strength, phase 

 

 

def detrend(data, axis=-1, type='linear', bp=0): 

""" 

Remove linear trend along axis from data. 

 

Parameters 

---------- 

data : array_like 

The input data. 

axis : int, optional 

The axis along which to detrend the data. By default this is the 

last axis (-1). 

type : {'linear', 'constant'}, optional 

The type of detrending. If ``type == 'linear'`` (default), 

the result of a linear least-squares fit to `data` is subtracted 

from `data`. 

If ``type == 'constant'``, only the mean of `data` is subtracted. 

bp : array_like of ints, optional 

A sequence of break points. If given, an individual linear fit is 

performed for each part of `data` between two break points. 

Break points are specified as indices into `data`. 

 

Returns 

------- 

ret : ndarray 

The detrended input data. 

 

Examples 

-------- 

>>> from scipy import signal 

>>> randgen = np.random.RandomState(9) 

>>> npoints = 1000 

>>> noise = randgen.randn(npoints) 

>>> x = 3 + 2*np.linspace(0, 1, npoints) + noise 

>>> (signal.detrend(x) - noise).max() < 0.01 

True 

 

""" 

if type not in ['linear', 'l', 'constant', 'c']: 

raise ValueError("Trend type must be 'linear' or 'constant'.") 

data = asarray(data) 

dtype = data.dtype.char 

if dtype not in 'dfDF': 

dtype = 'd' 

if type in ['constant', 'c']: 

ret = data - expand_dims(mean(data, axis), axis) 

return ret 

else: 

dshape = data.shape 

N = dshape[axis] 

bp = sort(unique(r_[0, bp, N])) 

if np.any(bp > N): 

raise ValueError("Breakpoints must be less than length " 

"of data along given axis.") 

Nreg = len(bp) - 1 

# Restructure data so that axis is along first dimension and 

# all other dimensions are collapsed into second dimension 

rnk = len(dshape) 

if axis < 0: 

axis = axis + rnk 

newdims = r_[axis, 0:axis, axis + 1:rnk] 

newdata = reshape(transpose(data, tuple(newdims)), 

(N, _prod(dshape) // N)) 

newdata = newdata.copy() # make sure we have a copy 

if newdata.dtype.char not in 'dfDF': 

newdata = newdata.astype(dtype) 

# Find leastsq fit and remove it for each piece 

for m in range(Nreg): 

Npts = bp[m + 1] - bp[m] 

A = ones((Npts, 2), dtype) 

A[:, 0] = cast[dtype](arange(1, Npts + 1) * 1.0 / Npts) 

sl = slice(bp[m], bp[m + 1]) 

coef, resids, rank, s = linalg.lstsq(A, newdata[sl]) 

newdata[sl] = newdata[sl] - dot(A, coef) 

# Put data back in original shape. 

tdshape = take(dshape, newdims, 0) 

ret = reshape(newdata, tuple(tdshape)) 

vals = list(range(1, rnk)) 

olddims = vals[:axis] + [0] + vals[axis:] 

ret = transpose(ret, tuple(olddims)) 

return ret 

 

 

def lfilter_zi(b, a): 

""" 

Construct initial conditions for lfilter for step response steady-state. 

 

Compute an initial state `zi` for the `lfilter` function that corresponds 

to the steady state of the step response. 

 

A typical use of this function is to set the initial state so that the 

output of the filter starts at the same value as the first element of 

the signal to be filtered. 

 

Parameters 

---------- 

b, a : array_like (1-D) 

The IIR filter coefficients. See `lfilter` for more 

information. 

 

Returns 

------- 

zi : 1-D ndarray 

The initial state for the filter. 

 

See Also 

-------- 

lfilter, lfiltic, filtfilt 

 

Notes 

----- 

A linear filter with order m has a state space representation (A, B, C, D), 

for which the output y of the filter can be expressed as:: 

 

z(n+1) = A*z(n) + B*x(n) 

y(n) = C*z(n) + D*x(n) 

 

where z(n) is a vector of length m, A has shape (m, m), B has shape 

(m, 1), C has shape (1, m) and D has shape (1, 1) (assuming x(n) is 

a scalar). lfilter_zi solves:: 

 

zi = A*zi + B 

 

In other words, it finds the initial condition for which the response 

to an input of all ones is a constant. 

 

Given the filter coefficients `a` and `b`, the state space matrices 

for the transposed direct form II implementation of the linear filter, 

which is the implementation used by scipy.signal.lfilter, are:: 

 

A = scipy.linalg.companion(a).T 

B = b[1:] - a[1:]*b[0] 

 

assuming `a[0]` is 1.0; if `a[0]` is not 1, `a` and `b` are first 

divided by a[0]. 

 

Examples 

-------- 

The following code creates a lowpass Butterworth filter. Then it 

applies that filter to an array whose values are all 1.0; the 

output is also all 1.0, as expected for a lowpass filter. If the 

`zi` argument of `lfilter` had not been given, the output would have 

shown the transient signal. 

 

>>> from numpy import array, ones 

>>> from scipy.signal import lfilter, lfilter_zi, butter 

>>> b, a = butter(5, 0.25) 

>>> zi = lfilter_zi(b, a) 

>>> y, zo = lfilter(b, a, ones(10), zi=zi) 

>>> y 

array([1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]) 

 

Another example: 

 

>>> x = array([0.5, 0.5, 0.5, 0.0, 0.0, 0.0, 0.0]) 

>>> y, zf = lfilter(b, a, x, zi=zi*x[0]) 

>>> y 

array([ 0.5 , 0.5 , 0.5 , 0.49836039, 0.48610528, 

0.44399389, 0.35505241]) 

 

Note that the `zi` argument to `lfilter` was computed using 

`lfilter_zi` and scaled by `x[0]`. Then the output `y` has no 

transient until the input drops from 0.5 to 0.0. 

 

""" 

 

# FIXME: Can this function be replaced with an appropriate 

# use of lfiltic? For example, when b,a = butter(N,Wn), 

# lfiltic(b, a, y=numpy.ones_like(a), x=numpy.ones_like(b)). 

# 

 

# We could use scipy.signal.normalize, but it uses warnings in 

# cases where a ValueError is more appropriate, and it allows 

# b to be 2D. 

b = np.atleast_1d(b) 

if b.ndim != 1: 

raise ValueError("Numerator b must be 1-D.") 

a = np.atleast_1d(a) 

if a.ndim != 1: 

raise ValueError("Denominator a must be 1-D.") 

 

while len(a) > 1 and a[0] == 0.0: 

a = a[1:] 

if a.size < 1: 

raise ValueError("There must be at least one nonzero `a` coefficient.") 

 

if a[0] != 1.0: 

# Normalize the coefficients so a[0] == 1. 

b = b / a[0] 

a = a / a[0] 

 

n = max(len(a), len(b)) 

 

# Pad a or b with zeros so they are the same length. 

if len(a) < n: 

a = np.r_[a, np.zeros(n - len(a))] 

elif len(b) < n: 

b = np.r_[b, np.zeros(n - len(b))] 

 

IminusA = np.eye(n - 1) - linalg.companion(a).T 

B = b[1:] - a[1:] * b[0] 

# Solve zi = A*zi + B 

zi = np.linalg.solve(IminusA, B) 

 

# For future reference: we could also use the following 

# explicit formulas to solve the linear system: 

# 

# zi = np.zeros(n - 1) 

# zi[0] = B.sum() / IminusA[:,0].sum() 

# asum = 1.0 

# csum = 0.0 

# for k in range(1,n-1): 

# asum += a[k] 

# csum += b[k] - a[k]*b[0] 

# zi[k] = asum*zi[0] - csum 

 

return zi 

 

 

def sosfilt_zi(sos): 

""" 

Construct initial conditions for sosfilt for step response steady-state. 

 

Compute an initial state `zi` for the `sosfilt` function that corresponds 

to the steady state of the step response. 

 

A typical use of this function is to set the initial state so that the 

output of the filter starts at the same value as the first element of 

the signal to be filtered. 

 

Parameters 

---------- 

sos : array_like 

Array of second-order filter coefficients, must have shape 

``(n_sections, 6)``. See `sosfilt` for the SOS filter format 

specification. 

 

Returns 

------- 

zi : ndarray 

Initial conditions suitable for use with ``sosfilt``, shape 

``(n_sections, 2)``. 

 

See Also 

-------- 

sosfilt, zpk2sos 

 

Notes 

----- 

.. versionadded:: 0.16.0 

 

Examples 

-------- 

Filter a rectangular pulse that begins at time 0, with and without 

the use of the `zi` argument of `scipy.signal.sosfilt`. 

 

>>> from scipy import signal 

>>> import matplotlib.pyplot as plt 

 

>>> sos = signal.butter(9, 0.125, output='sos') 

>>> zi = signal.sosfilt_zi(sos) 

>>> x = (np.arange(250) < 100).astype(int) 

>>> f1 = signal.sosfilt(sos, x) 

>>> f2, zo = signal.sosfilt(sos, x, zi=zi) 

 

>>> plt.plot(x, 'k--', label='x') 

>>> plt.plot(f1, 'b', alpha=0.5, linewidth=2, label='filtered') 

>>> plt.plot(f2, 'g', alpha=0.25, linewidth=4, label='filtered with zi') 

>>> plt.legend(loc='best') 

>>> plt.show() 

 

""" 

sos = np.asarray(sos) 

if sos.ndim != 2 or sos.shape[1] != 6: 

raise ValueError('sos must be shape (n_sections, 6)') 

 

n_sections = sos.shape[0] 

zi = np.empty((n_sections, 2)) 

scale = 1.0 

for section in range(n_sections): 

b = sos[section, :3] 

a = sos[section, 3:] 

zi[section] = scale * lfilter_zi(b, a) 

# If H(z) = B(z)/A(z) is this section's transfer function, then 

# b.sum()/a.sum() is H(1), the gain at omega=0. That's the steady 

# state value of this section's step response. 

scale *= b.sum() / a.sum() 

 

return zi 

 

 

def _filtfilt_gust(b, a, x, axis=-1, irlen=None): 

"""Forward-backward IIR filter that uses Gustafsson's method. 

 

Apply the IIR filter defined by `(b,a)` to `x` twice, first forward 

then backward, using Gustafsson's initial conditions [1]_. 

 

Let ``y_fb`` be the result of filtering first forward and then backward, 

and let ``y_bf`` be the result of filtering first backward then forward. 

Gustafsson's method is to compute initial conditions for the forward 

pass and the backward pass such that ``y_fb == y_bf``. 

 

Parameters 

---------- 

b : scalar or 1-D ndarray 

Numerator coefficients of the filter. 

a : scalar or 1-D ndarray 

Denominator coefficients of the filter. 

x : ndarray 

Data to be filtered. 

axis : int, optional 

Axis of `x` to be filtered. Default is -1. 

irlen : int or None, optional 

The length of the nonnegligible part of the impulse response. 

If `irlen` is None, or if the length of the signal is less than 

``2 * irlen``, then no part of the impulse response is ignored. 

 

Returns 

------- 

y : ndarray 

The filtered data. 

x0 : ndarray 

Initial condition for the forward filter. 

x1 : ndarray 

Initial condition for the backward filter. 

 

Notes 

----- 

Typically the return values `x0` and `x1` are not needed by the 

caller. The intended use of these return values is in unit tests. 

 

References 

---------- 

.. [1] F. Gustaffson. Determining the initial states in forward-backward 

filtering. Transactions on Signal Processing, 46(4):988-992, 1996. 

 

""" 

# In the comments, "Gustafsson's paper" and [1] refer to the 

# paper referenced in the docstring. 

 

b = np.atleast_1d(b) 

a = np.atleast_1d(a) 

 

order = max(len(b), len(a)) - 1 

if order == 0: 

# The filter is just scalar multiplication, with no state. 

scale = (b[0] / a[0])**2 

y = scale * x 

return y, np.array([]), np.array([]) 

 

if axis != -1 or axis != x.ndim - 1: 

# Move the axis containing the data to the end. 

x = np.swapaxes(x, axis, x.ndim - 1) 

 

# n is the number of samples in the data to be filtered. 

n = x.shape[-1] 

 

if irlen is None or n <= 2*irlen: 

m = n 

else: 

m = irlen 

 

# Create Obs, the observability matrix (called O in the paper). 

# This matrix can be interpreted as the operator that propagates 

# an arbitrary initial state to the output, assuming the input is 

# zero. 

# In Gustafsson's paper, the forward and backward filters are not 

# necessarily the same, so he has both O_f and O_b. We use the same 

# filter in both directions, so we only need O. The same comment 

# applies to S below. 

Obs = np.zeros((m, order)) 

zi = np.zeros(order) 

zi[0] = 1 

Obs[:, 0] = lfilter(b, a, np.zeros(m), zi=zi)[0] 

for k in range(1, order): 

Obs[k:, k] = Obs[:-k, 0] 

 

# Obsr is O^R (Gustafsson's notation for row-reversed O) 

Obsr = Obs[::-1] 

 

# Create S. S is the matrix that applies the filter to the reversed 

# propagated initial conditions. That is, 

# out = S.dot(zi) 

# is the same as 

# tmp, _ = lfilter(b, a, zeros(), zi=zi) # Propagate ICs. 

# out = lfilter(b, a, tmp[::-1]) # Reverse and filter. 

 

# Equations (5) & (6) of [1] 

S = lfilter(b, a, Obs[::-1], axis=0) 

 

# Sr is S^R (row-reversed S) 

Sr = S[::-1] 

 

# M is [(S^R - O), (O^R - S)] 

if m == n: 

M = np.hstack((Sr - Obs, Obsr - S)) 

else: 

# Matrix described in section IV of [1]. 

M = np.zeros((2*m, 2*order)) 

M[:m, :order] = Sr - Obs 

M[m:, order:] = Obsr - S 

 

# Naive forward-backward and backward-forward filters. 

# These have large transients because the filters use zero initial 

# conditions. 

y_f = lfilter(b, a, x) 

y_fb = lfilter(b, a, y_f[..., ::-1])[..., ::-1] 

 

y_b = lfilter(b, a, x[..., ::-1])[..., ::-1] 

y_bf = lfilter(b, a, y_b) 

 

delta_y_bf_fb = y_bf - y_fb 

if m == n: 

delta = delta_y_bf_fb 

else: 

start_m = delta_y_bf_fb[..., :m] 

end_m = delta_y_bf_fb[..., -m:] 

delta = np.concatenate((start_m, end_m), axis=-1) 

 

# ic_opt holds the "optimal" initial conditions. 

# The following code computes the result shown in the formula 

# of the paper between equations (6) and (7). 

if delta.ndim == 1: 

ic_opt = linalg.lstsq(M, delta)[0] 

else: 

# Reshape delta so it can be used as an array of multiple 

# right-hand-sides in linalg.lstsq. 

delta2d = delta.reshape(-1, delta.shape[-1]).T 

ic_opt0 = linalg.lstsq(M, delta2d)[0].T 

ic_opt = ic_opt0.reshape(delta.shape[:-1] + (M.shape[-1],)) 

 

# Now compute the filtered signal using equation (7) of [1]. 

# First, form [S^R, O^R] and call it W. 

if m == n: 

W = np.hstack((Sr, Obsr)) 

else: 

W = np.zeros((2*m, 2*order)) 

W[:m, :order] = Sr 

W[m:, order:] = Obsr 

 

# Equation (7) of [1] says 

# Y_fb^opt = Y_fb^0 + W * [x_0^opt; x_{N-1}^opt] 

# `wic` is (almost) the product on the right. 

# W has shape (m, 2*order), and ic_opt has shape (..., 2*order), 

# so we can't use W.dot(ic_opt). Instead, we dot ic_opt with W.T, 

# so wic has shape (..., m). 

wic = ic_opt.dot(W.T) 

 

# `wic` is "almost" the product of W and the optimal ICs in equation 

# (7)--if we're using a truncated impulse response (m < n), `wic` 

# contains only the adjustments required for the ends of the signal. 

# Here we form y_opt, taking this into account if necessary. 

y_opt = y_fb 

if m == n: 

y_opt += wic 

else: 

y_opt[..., :m] += wic[..., :m] 

y_opt[..., -m:] += wic[..., -m:] 

 

x0 = ic_opt[..., :order] 

x1 = ic_opt[..., -order:] 

if axis != -1 or axis != x.ndim - 1: 

# Restore the data axis to its original position. 

x0 = np.swapaxes(x0, axis, x.ndim - 1) 

x1 = np.swapaxes(x1, axis, x.ndim - 1) 

y_opt = np.swapaxes(y_opt, axis, x.ndim - 1) 

 

return y_opt, x0, x1 

 

 

def filtfilt(b, a, x, axis=-1, padtype='odd', padlen=None, method='pad', 

irlen=None): 

""" 

Apply a digital filter forward and backward to a signal. 

 

This function applies a linear digital filter twice, once forward and 

once backwards. The combined filter has zero phase and a filter order 

twice that of the original. 

 

The function provides options for handling the edges of the signal. 

 

Parameters 

---------- 

b : (N,) array_like 

The numerator coefficient vector of the filter. 

a : (N,) array_like 

The denominator coefficient vector of the filter. If ``a[0]`` 

is not 1, then both `a` and `b` are normalized by ``a[0]``. 

x : array_like 

The array of data to be filtered. 

axis : int, optional 

The axis of `x` to which the filter is applied. 

Default is -1. 

padtype : str or None, optional 

Must be 'odd', 'even', 'constant', or None. This determines the 

type of extension to use for the padded signal to which the filter 

is applied. If `padtype` is None, no padding is used. The default 

is 'odd'. 

padlen : int or None, optional 

The number of elements by which to extend `x` at both ends of 

`axis` before applying the filter. This value must be less than 

``x.shape[axis] - 1``. ``padlen=0`` implies no padding. 

The default value is ``3 * max(len(a), len(b))``. 

method : str, optional 

Determines the method for handling the edges of the signal, either 

"pad" or "gust". When `method` is "pad", the signal is padded; the 

type of padding is determined by `padtype` and `padlen`, and `irlen` 

is ignored. When `method` is "gust", Gustafsson's method is used, 

and `padtype` and `padlen` are ignored. 

irlen : int or None, optional 

When `method` is "gust", `irlen` specifies the length of the 

impulse response of the filter. If `irlen` is None, no part 

of the impulse response is ignored. For a long signal, specifying 

`irlen` can significantly improve the performance of the filter. 

 

Returns 

------- 

y : ndarray 

The filtered output with the same shape as `x`. 

 

See Also 

-------- 

sosfiltfilt, lfilter_zi, lfilter, lfiltic, savgol_filter, sosfilt 

 

Notes 

----- 

When `method` is "pad", the function pads the data along the given axis 

in one of three ways: odd, even or constant. The odd and even extensions 

have the corresponding symmetry about the end point of the data. The 

constant extension extends the data with the values at the end points. On 

both the forward and backward passes, the initial condition of the 

filter is found by using `lfilter_zi` and scaling it by the end point of 

the extended data. 

 

When `method` is "gust", Gustafsson's method [1]_ is used. Initial 

conditions are chosen for the forward and backward passes so that the 

forward-backward filter gives the same result as the backward-forward 

filter. 

 

The option to use Gustaffson's method was added in scipy version 0.16.0. 

 

References 

---------- 

.. [1] F. Gustaffson, "Determining the initial states in forward-backward 

filtering", Transactions on Signal Processing, Vol. 46, pp. 988-992, 

1996. 

 

Examples 

-------- 

The examples will use several functions from `scipy.signal`. 

 

>>> from scipy import signal 

>>> import matplotlib.pyplot as plt 

 

First we create a one second signal that is the sum of two pure sine 

waves, with frequencies 5 Hz and 250 Hz, sampled at 2000 Hz. 

 

>>> t = np.linspace(0, 1.0, 2001) 

>>> xlow = np.sin(2 * np.pi * 5 * t) 

>>> xhigh = np.sin(2 * np.pi * 250 * t) 

>>> x = xlow + xhigh 

 

Now create a lowpass Butterworth filter with a cutoff of 0.125 times 

the Nyquist frequency, or 125 Hz, and apply it to ``x`` with `filtfilt`. 

The result should be approximately ``xlow``, with no phase shift. 

 

>>> b, a = signal.butter(8, 0.125) 

>>> y = signal.filtfilt(b, a, x, padlen=150) 

>>> np.abs(y - xlow).max() 

9.1086182074789912e-06 

 

We get a fairly clean result for this artificial example because 

the odd extension is exact, and with the moderately long padding, 

the filter's transients have dissipated by the time the actual data 

is reached. In general, transient effects at the edges are 

unavoidable. 

 

The following example demonstrates the option ``method="gust"``. 

 

First, create a filter. 

 

>>> b, a = signal.ellip(4, 0.01, 120, 0.125) # Filter to be applied. 

>>> np.random.seed(123456) 

 

`sig` is a random input signal to be filtered. 

 

>>> n = 60 

>>> sig = np.random.randn(n)**3 + 3*np.random.randn(n).cumsum() 

 

Apply `filtfilt` to `sig`, once using the Gustafsson method, and 

once using padding, and plot the results for comparison. 

 

>>> fgust = signal.filtfilt(b, a, sig, method="gust") 

>>> fpad = signal.filtfilt(b, a, sig, padlen=50) 

>>> plt.plot(sig, 'k-', label='input') 

>>> plt.plot(fgust, 'b-', linewidth=4, label='gust') 

>>> plt.plot(fpad, 'c-', linewidth=1.5, label='pad') 

>>> plt.legend(loc='best') 

>>> plt.show() 

 

The `irlen` argument can be used to improve the performance 

of Gustafsson's method. 

 

Estimate the impulse response length of the filter. 

 

>>> z, p, k = signal.tf2zpk(b, a) 

>>> eps = 1e-9 

>>> r = np.max(np.abs(p)) 

>>> approx_impulse_len = int(np.ceil(np.log(eps) / np.log(r))) 

>>> approx_impulse_len 

137 

 

Apply the filter to a longer signal, with and without the `irlen` 

argument. The difference between `y1` and `y2` is small. For long 

signals, using `irlen` gives a significant performance improvement. 

 

>>> x = np.random.randn(5000) 

>>> y1 = signal.filtfilt(b, a, x, method='gust') 

>>> y2 = signal.filtfilt(b, a, x, method='gust', irlen=approx_impulse_len) 

>>> print(np.max(np.abs(y1 - y2))) 

1.80056858312e-10 

 

""" 

b = np.atleast_1d(b) 

a = np.atleast_1d(a) 

x = np.asarray(x) 

 

if method not in ["pad", "gust"]: 

raise ValueError("method must be 'pad' or 'gust'.") 

 

if method == "gust": 

y, z1, z2 = _filtfilt_gust(b, a, x, axis=axis, irlen=irlen) 

return y 

 

# method == "pad" 

edge, ext = _validate_pad(padtype, padlen, x, axis, 

ntaps=max(len(a), len(b))) 

 

# Get the steady state of the filter's step response. 

zi = lfilter_zi(b, a) 

 

# Reshape zi and create x0 so that zi*x0 broadcasts 

# to the correct value for the 'zi' keyword argument 

# to lfilter. 

zi_shape = [1] * x.ndim 

zi_shape[axis] = zi.size 

zi = np.reshape(zi, zi_shape) 

x0 = axis_slice(ext, stop=1, axis=axis) 

 

# Forward filter. 

(y, zf) = lfilter(b, a, ext, axis=axis, zi=zi * x0) 

 

# Backward filter. 

# Create y0 so zi*y0 broadcasts appropriately. 

y0 = axis_slice(y, start=-1, axis=axis) 

(y, zf) = lfilter(b, a, axis_reverse(y, axis=axis), axis=axis, zi=zi * y0) 

 

# Reverse y. 

y = axis_reverse(y, axis=axis) 

 

if edge > 0: 

# Slice the actual signal from the extended signal. 

y = axis_slice(y, start=edge, stop=-edge, axis=axis) 

 

return y 

 

 

def _validate_pad(padtype, padlen, x, axis, ntaps): 

"""Helper to validate padding for filtfilt""" 

if padtype not in ['even', 'odd', 'constant', None]: 

raise ValueError(("Unknown value '%s' given to padtype. padtype " 

"must be 'even', 'odd', 'constant', or None.") % 

padtype) 

 

if padtype is None: 

padlen = 0 

 

if padlen is None: 

# Original padding; preserved for backwards compatibility. 

edge = ntaps * 3 

else: 

edge = padlen 

 

# x's 'axis' dimension must be bigger than edge. 

if x.shape[axis] <= edge: 

raise ValueError("The length of the input vector x must be at least " 

"padlen, which is %d." % edge) 

 

if padtype is not None and edge > 0: 

# Make an extension of length `edge` at each 

# end of the input array. 

if padtype == 'even': 

ext = even_ext(x, edge, axis=axis) 

elif padtype == 'odd': 

ext = odd_ext(x, edge, axis=axis) 

else: 

ext = const_ext(x, edge, axis=axis) 

else: 

ext = x 

return edge, ext 

 

 

def sosfilt(sos, x, axis=-1, zi=None): 

""" 

Filter data along one dimension using cascaded second-order sections. 

 

Filter a data sequence, `x`, using a digital IIR filter defined by 

`sos`. This is implemented by performing `lfilter` for each 

second-order section. See `lfilter` for details. 

 

Parameters 

---------- 

sos : array_like 

Array of second-order filter coefficients, must have shape 

``(n_sections, 6)``. Each row corresponds to a second-order 

section, with the first three columns providing the numerator 

coefficients and the last three providing the denominator 

coefficients. 

x : array_like 

An N-dimensional input array. 

axis : int, optional 

The axis of the input data array along which to apply the 

linear filter. The filter is applied to each subarray along 

this axis. Default is -1. 

zi : array_like, optional 

Initial conditions for the cascaded filter delays. It is a (at 

least 2D) vector of shape ``(n_sections, ..., 2, ...)``, where 

``..., 2, ...`` denotes the shape of `x`, but with ``x.shape[axis]`` 

replaced by 2. If `zi` is None or is not given then initial rest 

(i.e. all zeros) is assumed. 

Note that these initial conditions are *not* the same as the initial 

conditions given by `lfiltic` or `lfilter_zi`. 

 

Returns 

------- 

y : ndarray 

The output of the digital filter. 

zf : ndarray, optional 

If `zi` is None, this is not returned, otherwise, `zf` holds the 

final filter delay values. 

 

See Also 

-------- 

zpk2sos, sos2zpk, sosfilt_zi, sosfiltfilt, sosfreqz 

 

Notes 

----- 

The filter function is implemented as a series of second-order filters 

with direct-form II transposed structure. It is designed to minimize 

numerical precision errors for high-order filters. 

 

.. versionadded:: 0.16.0 

 

Examples 

-------- 

Plot a 13th-order filter's impulse response using both `lfilter` and 

`sosfilt`, showing the instability that results from trying to do a 

13th-order filter in a single stage (the numerical error pushes some poles 

outside of the unit circle): 

 

>>> import matplotlib.pyplot as plt 

>>> from scipy import signal 

>>> b, a = signal.ellip(13, 0.009, 80, 0.05, output='ba') 

>>> sos = signal.ellip(13, 0.009, 80, 0.05, output='sos') 

>>> x = signal.unit_impulse(700) 

>>> y_tf = signal.lfilter(b, a, x) 

>>> y_sos = signal.sosfilt(sos, x) 

>>> plt.plot(y_tf, 'r', label='TF') 

>>> plt.plot(y_sos, 'k', label='SOS') 

>>> plt.legend(loc='best') 

>>> plt.show() 

 

""" 

x = np.asarray(x) 

sos, n_sections = _validate_sos(sos) 

use_zi = zi is not None 

if use_zi: 

zi = np.asarray(zi) 

x_zi_shape = list(x.shape) 

x_zi_shape[axis] = 2 

x_zi_shape = tuple([n_sections] + x_zi_shape) 

if zi.shape != x_zi_shape: 

raise ValueError('Invalid zi shape. With axis=%r, an input with ' 

'shape %r, and an sos array with %d sections, zi ' 

'must have shape %r, got %r.' % 

(axis, x.shape, n_sections, x_zi_shape, zi.shape)) 

zf = zeros_like(zi) 

 

for section in range(n_sections): 

if use_zi: 

x, zf[section] = lfilter(sos[section, :3], sos[section, 3:], 

x, axis, zi=zi[section]) 

else: 

x = lfilter(sos[section, :3], sos[section, 3:], x, axis) 

out = (x, zf) if use_zi else x 

return out 

 

 

def sosfiltfilt(sos, x, axis=-1, padtype='odd', padlen=None): 

""" 

A forward-backward digital filter using cascaded second-order sections. 

 

See `filtfilt` for more complete information about this method. 

 

Parameters 

---------- 

sos : array_like 

Array of second-order filter coefficients, must have shape 

``(n_sections, 6)``. Each row corresponds to a second-order 

section, with the first three columns providing the numerator 

coefficients and the last three providing the denominator 

coefficients. 

x : array_like 

The array of data to be filtered. 

axis : int, optional 

The axis of `x` to which the filter is applied. 

Default is -1. 

padtype : str or None, optional 

Must be 'odd', 'even', 'constant', or None. This determines the 

type of extension to use for the padded signal to which the filter 

is applied. If `padtype` is None, no padding is used. The default 

is 'odd'. 

padlen : int or None, optional 

The number of elements by which to extend `x` at both ends of 

`axis` before applying the filter. This value must be less than 

``x.shape[axis] - 1``. ``padlen=0`` implies no padding. 

The default value is:: 

 

3 * (2 * len(sos) + 1 - min((sos[:, 2] == 0).sum(), 

(sos[:, 5] == 0).sum())) 

 

The extra subtraction at the end attempts to compensate for poles 

and zeros at the origin (e.g. for odd-order filters) to yield 

equivalent estimates of `padlen` to those of `filtfilt` for 

second-order section filters built with `scipy.signal` functions. 

 

Returns 

------- 

y : ndarray 

The filtered output with the same shape as `x`. 

 

See Also 

-------- 

filtfilt, sosfilt, sosfilt_zi, sosfreqz 

 

Notes 

----- 

.. versionadded:: 0.18.0 

 

Examples 

-------- 

>>> from scipy.signal import sosfiltfilt, butter 

>>> import matplotlib.pyplot as plt 

 

Create an interesting signal to filter. 

 

>>> n = 201 

>>> t = np.linspace(0, 1, n) 

>>> np.random.seed(123) 

>>> x = 1 + (t < 0.5) - 0.25*t**2 + 0.05*np.random.randn(n) 

 

Create a lowpass Butterworth filter, and use it to filter `x`. 

 

>>> sos = butter(4, 0.125, output='sos') 

>>> y = sosfiltfilt(sos, x) 

 

For comparison, apply an 8th order filter using `sosfilt`. The filter 

is initialized using the mean of the first four values of `x`. 

 

>>> from scipy.signal import sosfilt, sosfilt_zi 

>>> sos8 = butter(8, 0.125, output='sos') 

>>> zi = x[:4].mean() * sosfilt_zi(sos8) 

>>> y2, zo = sosfilt(sos8, x, zi=zi) 

 

Plot the results. Note that the phase of `y` matches the input, while 

`y2` has a significant phase delay. 

 

>>> plt.plot(t, x, alpha=0.5, label='x(t)') 

>>> plt.plot(t, y, label='y(t)') 

>>> plt.plot(t, y2, label='y2(t)') 

>>> plt.legend(framealpha=1, shadow=True) 

>>> plt.grid(alpha=0.25) 

>>> plt.xlabel('t') 

>>> plt.show() 

 

""" 

sos, n_sections = _validate_sos(sos) 

 

# `method` is "pad"... 

ntaps = 2 * n_sections + 1 

ntaps -= min((sos[:, 2] == 0).sum(), (sos[:, 5] == 0).sum()) 

edge, ext = _validate_pad(padtype, padlen, x, axis, 

ntaps=ntaps) 

 

# These steps follow the same form as filtfilt with modifications 

zi = sosfilt_zi(sos) # shape (n_sections, 2) --> (n_sections, ..., 2, ...) 

zi_shape = [1] * x.ndim 

zi_shape[axis] = 2 

zi.shape = [n_sections] + zi_shape 

x_0 = axis_slice(ext, stop=1, axis=axis) 

(y, zf) = sosfilt(sos, ext, axis=axis, zi=zi * x_0) 

y_0 = axis_slice(y, start=-1, axis=axis) 

(y, zf) = sosfilt(sos, axis_reverse(y, axis=axis), axis=axis, zi=zi * y_0) 

y = axis_reverse(y, axis=axis) 

if edge > 0: 

y = axis_slice(y, start=edge, stop=-edge, axis=axis) 

return y 

 

 

def decimate(x, q, n=None, ftype='iir', axis=-1, zero_phase=True): 

""" 

Downsample the signal after applying an anti-aliasing filter. 

 

By default, an order 8 Chebyshev type I filter is used. A 30 point FIR 

filter with Hamming window is used if `ftype` is 'fir'. 

 

Parameters 

---------- 

x : array_like 

The signal to be downsampled, as an N-dimensional array. 

q : int 

The downsampling factor. When using IIR downsampling, it is recommended 

to call `decimate` multiple times for downsampling factors higher than 

13. 

n : int, optional 

The order of the filter (1 less than the length for 'fir'). Defaults to 

8 for 'iir' and 20 times the downsampling factor for 'fir'. 

ftype : str {'iir', 'fir'} or ``dlti`` instance, optional 

If 'iir' or 'fir', specifies the type of lowpass filter. If an instance 

of an `dlti` object, uses that object to filter before downsampling. 

axis : int, optional 

The axis along which to decimate. 

zero_phase : bool, optional 

Prevent phase shift by filtering with `filtfilt` instead of `lfilter` 

when using an IIR filter, and shifting the outputs back by the filter's 

group delay when using an FIR filter. The default value of ``True`` is 

recommended, since a phase shift is generally not desired. 

 

.. versionadded:: 0.18.0 

 

Returns 

------- 

y : ndarray 

The down-sampled signal. 

 

See Also 

-------- 

resample : Resample up or down using the FFT method. 

resample_poly : Resample using polyphase filtering and an FIR filter. 

 

Notes 

----- 

The ``zero_phase`` keyword was added in 0.18.0. 

The possibility to use instances of ``dlti`` as ``ftype`` was added in 

0.18.0. 

""" 

 

x = asarray(x) 

q = operator.index(q) 

 

if n is not None: 

n = operator.index(n) 

 

if ftype == 'fir': 

if n is None: 

half_len = 10 * q # reasonable cutoff for our sinc-like function 

n = 2 * half_len 

b, a = firwin(n+1, 1. / q, window='hamming'), 1. 

elif ftype == 'iir': 

if n is None: 

n = 8 

system = dlti(*cheby1(n, 0.05, 0.8 / q)) 

b, a = system.num, system.den 

elif isinstance(ftype, dlti): 

system = ftype._as_tf() # Avoids copying if already in TF form 

b, a = system.num, system.den 

else: 

raise ValueError('invalid ftype') 

 

sl = [slice(None)] * x.ndim 

a = np.asarray(a) 

 

if a.size == 1: # FIR case 

b = b / a 

if zero_phase: 

y = resample_poly(x, 1, q, axis=axis, window=b) 

else: 

# upfirdn is generally faster than lfilter by a factor equal to the 

# downsampling factor, since it only calculates the needed outputs 

n_out = x.shape[axis] // q + bool(x.shape[axis] % q) 

y = upfirdn(b, x, up=1, down=q, axis=axis) 

sl[axis] = slice(None, n_out, None) 

 

else: # IIR case 

if zero_phase: 

y = filtfilt(b, a, x, axis=axis) 

else: 

y = lfilter(b, a, x, axis=axis) 

sl[axis] = slice(None, None, q) 

 

return y[sl]