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

"""Tools for spectral analysis. 

""" 

 

from __future__ import division, print_function, absolute_import 

 

import numpy as np 

from scipy import fftpack 

from . import signaltools 

from .windows import get_window 

from ._spectral import _lombscargle 

from ._arraytools import const_ext, even_ext, odd_ext, zero_ext 

import warnings 

 

from scipy._lib.six import string_types 

 

__all__ = ['periodogram', 'welch', 'lombscargle', 'csd', 'coherence', 

'spectrogram', 'stft', 'istft', 'check_COLA'] 

 

 

def lombscargle(x, 

y, 

freqs, 

precenter=False, 

normalize=False): 

""" 

lombscargle(x, y, freqs) 

 

Computes the Lomb-Scargle periodogram. 

 

The Lomb-Scargle periodogram was developed by Lomb [1]_ and further 

extended by Scargle [2]_ to find, and test the significance of weak 

periodic signals with uneven temporal sampling. 

 

When *normalize* is False (default) the computed periodogram 

is unnormalized, it takes the value ``(A**2) * N/4`` for a harmonic 

signal with amplitude A for sufficiently large N. 

 

When *normalize* is True the computed periodogram is is normalized by 

the residuals of the data around a constant reference model (at zero). 

 

Input arrays should be one-dimensional and will be cast to float64. 

 

Parameters 

---------- 

x : array_like 

Sample times. 

y : array_like 

Measurement values. 

freqs : array_like 

Angular frequencies for output periodogram. 

precenter : bool, optional 

Pre-center amplitudes by subtracting the mean. 

normalize : bool, optional 

Compute normalized periodogram. 

 

Returns 

------- 

pgram : array_like 

Lomb-Scargle periodogram. 

 

Raises 

------ 

ValueError 

If the input arrays `x` and `y` do not have the same shape. 

 

Notes 

----- 

This subroutine calculates the periodogram using a slightly 

modified algorithm due to Townsend [3]_ which allows the 

periodogram to be calculated using only a single pass through 

the input arrays for each frequency. 

 

The algorithm running time scales roughly as O(x * freqs) or O(N^2) 

for a large number of samples and frequencies. 

 

References 

---------- 

.. [1] N.R. Lomb "Least-squares frequency analysis of unequally spaced 

data", Astrophysics and Space Science, vol 39, pp. 447-462, 1976 

 

.. [2] J.D. Scargle "Studies in astronomical time series analysis. II -  

Statistical aspects of spectral analysis of unevenly spaced data", 

The Astrophysical Journal, vol 263, pp. 835-853, 1982 

 

.. [3] R.H.D. Townsend, "Fast calculation of the Lomb-Scargle 

periodogram using graphics processing units.", The Astrophysical 

Journal Supplement Series, vol 191, pp. 247-253, 2010 

 

Examples 

-------- 

>>> import scipy.signal 

>>> import matplotlib.pyplot as plt 

 

First define some input parameters for the signal: 

 

>>> A = 2. 

>>> w = 1. 

>>> phi = 0.5 * np.pi 

>>> nin = 1000 

>>> nout = 100000 

>>> frac_points = 0.9 # Fraction of points to select 

 

Randomly select a fraction of an array with timesteps: 

 

>>> r = np.random.rand(nin) 

>>> x = np.linspace(0.01, 10*np.pi, nin) 

>>> x = x[r >= frac_points] 

 

Plot a sine wave for the selected times: 

 

>>> y = A * np.sin(w*x+phi) 

 

Define the array of frequencies for which to compute the periodogram: 

 

>>> f = np.linspace(0.01, 10, nout) 

 

Calculate Lomb-Scargle periodogram: 

 

>>> import scipy.signal as signal 

>>> pgram = signal.lombscargle(x, y, f, normalize=True) 

 

Now make a plot of the input data: 

 

>>> plt.subplot(2, 1, 1) 

>>> plt.plot(x, y, 'b+') 

 

Then plot the normalized periodogram: 

 

>>> plt.subplot(2, 1, 2) 

>>> plt.plot(f, pgram) 

>>> plt.show() 

 

""" 

 

x = np.asarray(x, dtype=np.float64) 

y = np.asarray(y, dtype=np.float64) 

freqs = np.asarray(freqs, dtype=np.float64) 

 

assert x.ndim == 1 

assert y.ndim == 1 

assert freqs.ndim == 1 

 

if precenter: 

pgram = _lombscargle(x, y - y.mean(), freqs) 

else: 

pgram = _lombscargle(x, y, freqs) 

 

if normalize: 

pgram *= 2 / np.dot(y, y) 

 

return pgram 

 

 

def periodogram(x, fs=1.0, window='boxcar', nfft=None, detrend='constant', 

return_onesided=True, scaling='density', axis=-1): 

""" 

Estimate power spectral density using a periodogram. 

 

Parameters 

---------- 

x : array_like 

Time series of measurement values 

fs : float, optional 

Sampling frequency of the `x` time series. Defaults to 1.0. 

window : str or tuple or array_like, optional 

Desired window to use. If `window` is a string or tuple, it is 

passed to `get_window` to generate the window values, which are 

DFT-even by default. See `get_window` for a list of windows and 

required parameters. If `window` is array_like it will be used 

directly as the window and its length must be nperseg. Defaults 

to 'boxcar'. 

nfft : int, optional 

Length of the FFT used. If `None` the length of `x` will be 

used. 

detrend : str or function or `False`, optional 

Specifies how to detrend each segment. If `detrend` is a 

string, it is passed as the `type` argument to the `detrend` 

function. If it is a function, it takes a segment and returns a 

detrended segment. If `detrend` is `False`, no detrending is 

done. Defaults to 'constant'. 

return_onesided : bool, optional 

If `True`, return a one-sided spectrum for real data. If 

`False` return a two-sided spectrum. Note that for complex 

data, a two-sided spectrum is always returned. 

scaling : { 'density', 'spectrum' }, optional 

Selects between computing the power spectral density ('density') 

where `Pxx` has units of V**2/Hz and computing the power 

spectrum ('spectrum') where `Pxx` has units of V**2, if `x` 

is measured in V and `fs` is measured in Hz. Defaults to 

'density' 

axis : int, optional 

Axis along which the periodogram is computed; the default is 

over the last axis (i.e. ``axis=-1``). 

 

Returns 

------- 

f : ndarray 

Array of sample frequencies. 

Pxx : ndarray 

Power spectral density or power spectrum of `x`. 

 

Notes 

----- 

.. versionadded:: 0.12.0 

 

See Also 

-------- 

welch: Estimate power spectral density using Welch's method 

lombscargle: Lomb-Scargle periodogram for unevenly sampled data 

 

Examples 

-------- 

>>> from scipy import signal 

>>> import matplotlib.pyplot as plt 

>>> np.random.seed(1234) 

 

Generate a test signal, a 2 Vrms sine wave at 1234 Hz, corrupted by 

0.001 V**2/Hz of white noise sampled at 10 kHz. 

 

>>> fs = 10e3 

>>> N = 1e5 

>>> amp = 2*np.sqrt(2) 

>>> freq = 1234.0 

>>> noise_power = 0.001 * fs / 2 

>>> time = np.arange(N) / fs 

>>> x = amp*np.sin(2*np.pi*freq*time) 

>>> x += np.random.normal(scale=np.sqrt(noise_power), size=time.shape) 

 

Compute and plot the power spectral density. 

 

>>> f, Pxx_den = signal.periodogram(x, fs) 

>>> plt.semilogy(f, Pxx_den) 

>>> plt.ylim([1e-7, 1e2]) 

>>> plt.xlabel('frequency [Hz]') 

>>> plt.ylabel('PSD [V**2/Hz]') 

>>> plt.show() 

 

If we average the last half of the spectral density, to exclude the 

peak, we can recover the noise power on the signal. 

 

>>> np.mean(Pxx_den[25000:]) 

0.00099728892368242854 

 

Now compute and plot the power spectrum. 

 

>>> f, Pxx_spec = signal.periodogram(x, fs, 'flattop', scaling='spectrum') 

>>> plt.figure() 

>>> plt.semilogy(f, np.sqrt(Pxx_spec)) 

>>> plt.ylim([1e-4, 1e1]) 

>>> plt.xlabel('frequency [Hz]') 

>>> plt.ylabel('Linear spectrum [V RMS]') 

>>> plt.show() 

 

The peak height in the power spectrum is an estimate of the RMS 

amplitude. 

 

>>> np.sqrt(Pxx_spec.max()) 

2.0077340678640727 

 

""" 

x = np.asarray(x) 

 

if x.size == 0: 

return np.empty(x.shape), np.empty(x.shape) 

 

if window is None: 

window = 'boxcar' 

 

if nfft is None: 

nperseg = x.shape[axis] 

elif nfft == x.shape[axis]: 

nperseg = nfft 

elif nfft > x.shape[axis]: 

nperseg = x.shape[axis] 

elif nfft < x.shape[axis]: 

s = [np.s_[:]]*len(x.shape) 

s[axis] = np.s_[:nfft] 

x = x[s] 

nperseg = nfft 

nfft = None 

 

return welch(x, fs, window, nperseg, 0, nfft, detrend, return_onesided, 

scaling, axis) 

 

 

def welch(x, fs=1.0, window='hann', nperseg=None, noverlap=None, nfft=None, 

detrend='constant', return_onesided=True, scaling='density', 

axis=-1): 

r""" 

Estimate power spectral density using Welch's method. 

 

Welch's method [1]_ computes an estimate of the power spectral 

density by dividing the data into overlapping segments, computing a 

modified periodogram for each segment and averaging the 

periodograms. 

 

Parameters 

---------- 

x : array_like 

Time series of measurement values 

fs : float, optional 

Sampling frequency of the `x` time series. Defaults to 1.0. 

window : str or tuple or array_like, optional 

Desired window to use. If `window` is a string or tuple, it is 

passed to `get_window` to generate the window values, which are 

DFT-even by default. See `get_window` for a list of windows and 

required parameters. If `window` is array_like it will be used 

directly as the window and its length must be nperseg. Defaults 

to a Hann window. 

nperseg : int, optional 

Length of each segment. Defaults to None, but if window is str or 

tuple, is set to 256, and if window is array_like, is set to the 

length of the window. 

noverlap : int, optional 

Number of points to overlap between segments. If `None`, 

``noverlap = nperseg // 2``. Defaults to `None`. 

nfft : int, optional 

Length of the FFT used, if a zero padded FFT is desired. If 

`None`, the FFT length is `nperseg`. Defaults to `None`. 

detrend : str or function or `False`, optional 

Specifies how to detrend each segment. If `detrend` is a 

string, it is passed as the `type` argument to the `detrend` 

function. If it is a function, it takes a segment and returns a 

detrended segment. If `detrend` is `False`, no detrending is 

done. Defaults to 'constant'. 

return_onesided : bool, optional 

If `True`, return a one-sided spectrum for real data. If 

`False` return a two-sided spectrum. Note that for complex 

data, a two-sided spectrum is always returned. 

scaling : { 'density', 'spectrum' }, optional 

Selects between computing the power spectral density ('density') 

where `Pxx` has units of V**2/Hz and computing the power 

spectrum ('spectrum') where `Pxx` has units of V**2, if `x` 

is measured in V and `fs` is measured in Hz. Defaults to 

'density' 

axis : int, optional 

Axis along which the periodogram is computed; the default is 

over the last axis (i.e. ``axis=-1``). 

 

Returns 

------- 

f : ndarray 

Array of sample frequencies. 

Pxx : ndarray 

Power spectral density or power spectrum of x. 

 

See Also 

-------- 

periodogram: Simple, optionally modified periodogram 

lombscargle: Lomb-Scargle periodogram for unevenly sampled data 

 

Notes 

----- 

An appropriate amount of overlap will depend on the choice of window 

and on your requirements. For the default Hann window an overlap of 

50% is a reasonable trade off between accurately estimating the 

signal power, while not over counting any of the data. Narrower 

windows may require a larger overlap. 

 

If `noverlap` is 0, this method is equivalent to Bartlett's method 

[2]_. 

 

.. versionadded:: 0.12.0 

 

References 

---------- 

.. [1] P. Welch, "The use of the fast Fourier transform for the 

estimation of power spectra: A method based on time averaging 

over short, modified periodograms", IEEE Trans. Audio 

Electroacoust. vol. 15, pp. 70-73, 1967. 

.. [2] M.S. Bartlett, "Periodogram Analysis and Continuous Spectra", 

Biometrika, vol. 37, pp. 1-16, 1950. 

 

Examples 

-------- 

>>> from scipy import signal 

>>> import matplotlib.pyplot as plt 

>>> np.random.seed(1234) 

 

Generate a test signal, a 2 Vrms sine wave at 1234 Hz, corrupted by 

0.001 V**2/Hz of white noise sampled at 10 kHz. 

 

>>> fs = 10e3 

>>> N = 1e5 

>>> amp = 2*np.sqrt(2) 

>>> freq = 1234.0 

>>> noise_power = 0.001 * fs / 2 

>>> time = np.arange(N) / fs 

>>> x = amp*np.sin(2*np.pi*freq*time) 

>>> x += np.random.normal(scale=np.sqrt(noise_power), size=time.shape) 

 

Compute and plot the power spectral density. 

 

>>> f, Pxx_den = signal.welch(x, fs, nperseg=1024) 

>>> plt.semilogy(f, Pxx_den) 

>>> plt.ylim([0.5e-3, 1]) 

>>> plt.xlabel('frequency [Hz]') 

>>> plt.ylabel('PSD [V**2/Hz]') 

>>> plt.show() 

 

If we average the last half of the spectral density, to exclude the 

peak, we can recover the noise power on the signal. 

 

>>> np.mean(Pxx_den[256:]) 

0.0009924865443739191 

 

Now compute and plot the power spectrum. 

 

>>> f, Pxx_spec = signal.welch(x, fs, 'flattop', 1024, scaling='spectrum') 

>>> plt.figure() 

>>> plt.semilogy(f, np.sqrt(Pxx_spec)) 

>>> plt.xlabel('frequency [Hz]') 

>>> plt.ylabel('Linear spectrum [V RMS]') 

>>> plt.show() 

 

The peak height in the power spectrum is an estimate of the RMS 

amplitude. 

 

>>> np.sqrt(Pxx_spec.max()) 

2.0077340678640727 

 

""" 

 

freqs, Pxx = csd(x, x, fs, window, nperseg, noverlap, nfft, detrend, 

return_onesided, scaling, axis) 

 

return freqs, Pxx.real 

 

 

def csd(x, y, fs=1.0, window='hann', nperseg=None, noverlap=None, nfft=None, 

detrend='constant', return_onesided=True, scaling='density', axis=-1): 

r""" 

Estimate the cross power spectral density, Pxy, using Welch's 

method. 

 

Parameters 

---------- 

x : array_like 

Time series of measurement values 

y : array_like 

Time series of measurement values 

fs : float, optional 

Sampling frequency of the `x` and `y` time series. Defaults 

to 1.0. 

window : str or tuple or array_like, optional 

Desired window to use. If `window` is a string or tuple, it is 

passed to `get_window` to generate the window values, which are 

DFT-even by default. See `get_window` for a list of windows and 

required parameters. If `window` is array_like it will be used 

directly as the window and its length must be nperseg. Defaults 

to a Hann window. 

nperseg : int, optional 

Length of each segment. Defaults to None, but if window is str or 

tuple, is set to 256, and if window is array_like, is set to the 

length of the window. 

noverlap: int, optional 

Number of points to overlap between segments. If `None`, 

``noverlap = nperseg // 2``. Defaults to `None`. 

nfft : int, optional 

Length of the FFT used, if a zero padded FFT is desired. If 

`None`, the FFT length is `nperseg`. Defaults to `None`. 

detrend : str or function or `False`, optional 

Specifies how to detrend each segment. If `detrend` is a 

string, it is passed as the `type` argument to the `detrend` 

function. If it is a function, it takes a segment and returns a 

detrended segment. If `detrend` is `False`, no detrending is 

done. Defaults to 'constant'. 

return_onesided : bool, optional 

If `True`, return a one-sided spectrum for real data. If 

`False` return a two-sided spectrum. Note that for complex 

data, a two-sided spectrum is always returned. 

scaling : { 'density', 'spectrum' }, optional 

Selects between computing the cross spectral density ('density') 

where `Pxy` has units of V**2/Hz and computing the cross spectrum 

('spectrum') where `Pxy` has units of V**2, if `x` and `y` are 

measured in V and `fs` is measured in Hz. Defaults to 'density' 

axis : int, optional 

Axis along which the CSD is computed for both inputs; the 

default is over the last axis (i.e. ``axis=-1``). 

 

Returns 

------- 

f : ndarray 

Array of sample frequencies. 

Pxy : ndarray 

Cross spectral density or cross power spectrum of x,y. 

 

See Also 

-------- 

periodogram: Simple, optionally modified periodogram 

lombscargle: Lomb-Scargle periodogram for unevenly sampled data 

welch: Power spectral density by Welch's method. [Equivalent to 

csd(x,x)] 

coherence: Magnitude squared coherence by Welch's method. 

 

Notes 

-------- 

By convention, Pxy is computed with the conjugate FFT of X 

multiplied by the FFT of Y. 

 

If the input series differ in length, the shorter series will be 

zero-padded to match. 

 

An appropriate amount of overlap will depend on the choice of window 

and on your requirements. For the default Hann window an overlap of 

50% is a reasonable trade off between accurately estimating the 

signal power, while not over counting any of the data. Narrower 

windows may require a larger overlap. 

 

.. versionadded:: 0.16.0 

 

References 

---------- 

.. [1] P. Welch, "The use of the fast Fourier transform for the 

estimation of power spectra: A method based on time averaging 

over short, modified periodograms", IEEE Trans. Audio 

Electroacoust. vol. 15, pp. 70-73, 1967. 

.. [2] Rabiner, Lawrence R., and B. Gold. "Theory and Application of 

Digital Signal Processing" Prentice-Hall, pp. 414-419, 1975 

 

Examples 

-------- 

>>> from scipy import signal 

>>> import matplotlib.pyplot as plt 

 

Generate two test signals with some common features. 

 

>>> fs = 10e3 

>>> N = 1e5 

>>> amp = 20 

>>> freq = 1234.0 

>>> noise_power = 0.001 * fs / 2 

>>> time = np.arange(N) / fs 

>>> b, a = signal.butter(2, 0.25, 'low') 

>>> x = np.random.normal(scale=np.sqrt(noise_power), size=time.shape) 

>>> y = signal.lfilter(b, a, x) 

>>> x += amp*np.sin(2*np.pi*freq*time) 

>>> y += np.random.normal(scale=0.1*np.sqrt(noise_power), size=time.shape) 

 

Compute and plot the magnitude of the cross spectral density. 

 

>>> f, Pxy = signal.csd(x, y, fs, nperseg=1024) 

>>> plt.semilogy(f, np.abs(Pxy)) 

>>> plt.xlabel('frequency [Hz]') 

>>> plt.ylabel('CSD [V**2/Hz]') 

>>> plt.show() 

""" 

 

freqs, _, Pxy = _spectral_helper(x, y, fs, window, nperseg, noverlap, nfft, 

detrend, return_onesided, scaling, axis, 

mode='psd') 

 

# Average over windows. 

if len(Pxy.shape) >= 2 and Pxy.size > 0: 

if Pxy.shape[-1] > 1: 

Pxy = Pxy.mean(axis=-1) 

else: 

Pxy = np.reshape(Pxy, Pxy.shape[:-1]) 

 

return freqs, Pxy 

 

 

def spectrogram(x, fs=1.0, window=('tukey',.25), nperseg=None, noverlap=None, 

nfft=None, detrend='constant', return_onesided=True, 

scaling='density', axis=-1, mode='psd'): 

""" 

Compute a spectrogram with consecutive Fourier transforms. 

 

Spectrograms can be used as a way of visualizing the change of a 

nonstationary signal's frequency content over time. 

 

Parameters 

---------- 

x : array_like 

Time series of measurement values 

fs : float, optional 

Sampling frequency of the `x` time series. Defaults to 1.0. 

window : str or tuple or array_like, optional 

Desired window to use. If `window` is a string or tuple, it is 

passed to `get_window` to generate the window values, which are 

DFT-even by default. See `get_window` for a list of windows and 

required parameters. If `window` is array_like it will be used 

directly as the window and its length must be nperseg. 

Defaults to a Tukey window with shape parameter of 0.25. 

nperseg : int, optional 

Length of each segment. Defaults to None, but if window is str or 

tuple, is set to 256, and if window is array_like, is set to the 

length of the window. 

noverlap : int, optional 

Number of points to overlap between segments. If `None`, 

``noverlap = nperseg // 8``. Defaults to `None`. 

nfft : int, optional 

Length of the FFT used, if a zero padded FFT is desired. If 

`None`, the FFT length is `nperseg`. Defaults to `None`. 

detrend : str or function or `False`, optional 

Specifies how to detrend each segment. If `detrend` is a 

string, it is passed as the `type` argument to the `detrend` 

function. If it is a function, it takes a segment and returns a 

detrended segment. If `detrend` is `False`, no detrending is 

done. Defaults to 'constant'. 

return_onesided : bool, optional 

If `True`, return a one-sided spectrum for real data. If 

`False` return a two-sided spectrum. Note that for complex 

data, a two-sided spectrum is always returned. 

scaling : { 'density', 'spectrum' }, optional 

Selects between computing the power spectral density ('density') 

where `Sxx` has units of V**2/Hz and computing the power 

spectrum ('spectrum') where `Sxx` has units of V**2, if `x` 

is measured in V and `fs` is measured in Hz. Defaults to 

'density'. 

axis : int, optional 

Axis along which the spectrogram is computed; the default is over 

the last axis (i.e. ``axis=-1``). 

mode : str, optional 

Defines what kind of return values are expected. Options are 

['psd', 'complex', 'magnitude', 'angle', 'phase']. 'complex' is 

equivalent to the output of `stft` with no padding or boundary 

extension. 'magnitude' returns the absolute magnitude of the 

STFT. 'angle' and 'phase' return the complex angle of the STFT, 

with and without unwrapping, respectively. 

 

Returns 

------- 

f : ndarray 

Array of sample frequencies. 

t : ndarray 

Array of segment times. 

Sxx : ndarray 

Spectrogram of x. By default, the last axis of Sxx corresponds 

to the segment times. 

 

See Also 

-------- 

periodogram: Simple, optionally modified periodogram 

lombscargle: Lomb-Scargle periodogram for unevenly sampled data 

welch: Power spectral density by Welch's method. 

csd: Cross spectral density by Welch's method. 

 

Notes 

----- 

An appropriate amount of overlap will depend on the choice of window 

and on your requirements. In contrast to welch's method, where the 

entire data stream is averaged over, one may wish to use a smaller 

overlap (or perhaps none at all) when computing a spectrogram, to 

maintain some statistical independence between individual segments. 

It is for this reason that the default window is a Tukey window with 

1/8th of a window's length overlap at each end. 

 

.. versionadded:: 0.16.0 

 

References 

---------- 

.. [1] Oppenheim, Alan V., Ronald W. Schafer, John R. Buck 

"Discrete-Time Signal Processing", Prentice Hall, 1999. 

 

Examples 

-------- 

>>> from scipy import signal 

>>> import matplotlib.pyplot as plt 

 

Generate a test signal, a 2 Vrms sine wave whose frequency is slowly 

modulated around 3kHz, corrupted by white noise of exponentially 

decreasing magnitude sampled at 10 kHz. 

 

>>> fs = 10e3 

>>> N = 1e5 

>>> amp = 2 * np.sqrt(2) 

>>> noise_power = 0.01 * fs / 2 

>>> time = np.arange(N) / float(fs) 

>>> mod = 500*np.cos(2*np.pi*0.25*time) 

>>> carrier = amp * np.sin(2*np.pi*3e3*time + mod) 

>>> noise = np.random.normal(scale=np.sqrt(noise_power), size=time.shape) 

>>> noise *= np.exp(-time/5) 

>>> x = carrier + noise 

 

Compute and plot the spectrogram. 

 

>>> f, t, Sxx = signal.spectrogram(x, fs) 

>>> plt.pcolormesh(t, f, Sxx) 

>>> plt.ylabel('Frequency [Hz]') 

>>> plt.xlabel('Time [sec]') 

>>> plt.show() 

""" 

modelist = ['psd', 'complex', 'magnitude', 'angle', 'phase'] 

if mode not in modelist: 

raise ValueError('unknown value for mode {}, must be one of {}' 

.format(mode, modelist)) 

 

# need to set default for nperseg before setting default for noverlap below 

window, nperseg = _triage_segments(window, nperseg, 

input_length=x.shape[axis]) 

 

# Less overlap than welch, so samples are more statisically independent 

if noverlap is None: 

noverlap = nperseg // 8 

 

if mode == 'psd': 

freqs, time, Sxx = _spectral_helper(x, x, fs, window, nperseg, 

noverlap, nfft, detrend, 

return_onesided, scaling, axis, 

mode='psd') 

 

else: 

freqs, time, Sxx = _spectral_helper(x, x, fs, window, nperseg, 

noverlap, nfft, detrend, 

return_onesided, scaling, axis, 

mode='stft') 

 

if mode == 'magnitude': 

Sxx = np.abs(Sxx) 

elif mode in ['angle', 'phase']: 

Sxx = np.angle(Sxx) 

if mode == 'phase': 

# Sxx has one additional dimension for time strides 

if axis < 0: 

axis -= 1 

Sxx = np.unwrap(Sxx, axis=axis) 

 

# mode =='complex' is same as `stft`, doesn't need modification 

 

return freqs, time, Sxx 

 

 

def check_COLA(window, nperseg, noverlap, tol=1e-10): 

r""" 

Check whether the Constant OverLap Add (COLA) constraint is met 

 

Parameters 

---------- 

window : str or tuple or array_like 

Desired window to use. If `window` is a string or tuple, it is 

passed to `get_window` to generate the window values, which are 

DFT-even by default. See `get_window` for a list of windows and 

required parameters. If `window` is array_like it will be used 

directly as the window and its length must be nperseg. 

nperseg : int 

Length of each segment. 

noverlap : int 

Number of points to overlap between segments. 

tol : float, optional 

The allowed variance of a bin's weighted sum from the median bin 

sum. 

 

Returns 

------- 

verdict : bool 

`True` if chosen combination satisfies COLA within `tol`, 

`False` otherwise 

 

See Also 

-------- 

stft: Short Time Fourier Transform 

istft: Inverse Short Time Fourier Transform 

 

Notes 

----- 

In order to enable inversion of an STFT via the inverse STFT in 

`istft`, the signal windowing must obey the constraint of "Constant 

OverLap Add" (COLA). This ensures that every point in the input data 

is equally weighted, thereby avoiding aliasing and allowing full 

reconstruction. 

 

Some examples of windows that satisfy COLA: 

- Rectangular window at overlap of 0, 1/2, 2/3, 3/4, ... 

- Bartlett window at overlap of 1/2, 3/4, 5/6, ... 

- Hann window at 1/2, 2/3, 3/4, ... 

- Any Blackman family window at 2/3 overlap 

- Any window with ``noverlap = nperseg-1`` 

 

A very comprehensive list of other windows may be found in [2]_, 

wherein the COLA condition is satisfied when the "Amplitude 

Flatness" is unity. 

 

.. versionadded:: 0.19.0 

 

References 

---------- 

.. [1] Julius O. Smith III, "Spectral Audio Signal Processing", W3K 

Publishing, 2011,ISBN 978-0-9745607-3-1. 

.. [2] G. Heinzel, A. Ruediger and R. Schilling, "Spectrum and 

spectral density estimation by the Discrete Fourier transform 

(DFT), including a comprehensive list of window functions and 

some new at-top windows", 2002, 

http://hdl.handle.net/11858/00-001M-0000-0013-557A-5 

 

Examples 

-------- 

>>> from scipy import signal 

 

Confirm COLA condition for rectangular window of 75% (3/4) overlap: 

 

>>> signal.check_COLA(signal.boxcar(100), 100, 75) 

True 

 

COLA is not true for 25% (1/4) overlap, though: 

 

>>> signal.check_COLA(signal.boxcar(100), 100, 25) 

False 

 

"Symmetrical" Hann window (for filter design) is not COLA: 

 

>>> signal.check_COLA(signal.hann(120, sym=True), 120, 60) 

False 

 

"Periodic" or "DFT-even" Hann window (for FFT analysis) is COLA for 

overlap of 1/2, 2/3, 3/4, etc.: 

 

>>> signal.check_COLA(signal.hann(120, sym=False), 120, 60) 

True 

 

>>> signal.check_COLA(signal.hann(120, sym=False), 120, 80) 

True 

 

>>> signal.check_COLA(signal.hann(120, sym=False), 120, 90) 

True 

 

""" 

 

nperseg = int(nperseg) 

 

if nperseg < 1: 

raise ValueError('nperseg must be a positive integer') 

 

if noverlap >= nperseg: 

raise ValueError('noverlap must be less than nperseg.') 

noverlap = int(noverlap) 

 

if isinstance(window, string_types) or type(window) is tuple: 

win = get_window(window, nperseg) 

else: 

win = np.asarray(window) 

if len(win.shape) != 1: 

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

if win.shape[0] != nperseg: 

raise ValueError('window must have length of nperseg') 

 

step = nperseg - noverlap 

binsums = sum(win[ii*step:(ii+1)*step] for ii in range(nperseg//step)) 

 

if nperseg % step != 0: 

binsums[:nperseg % step] += win[-(nperseg % step):] 

 

deviation = binsums - np.median(binsums) 

return np.max(np.abs(deviation)) < tol 

 

 

def stft(x, fs=1.0, window='hann', nperseg=256, noverlap=None, nfft=None, 

detrend=False, return_onesided=True, boundary='zeros', padded=True, 

axis=-1): 

r""" 

Compute the Short Time Fourier Transform (STFT). 

 

STFTs can be used as a way of quantifying the change of a 

nonstationary signal's frequency and phase content over time. 

 

Parameters 

---------- 

x : array_like 

Time series of measurement values 

fs : float, optional 

Sampling frequency of the `x` time series. Defaults to 1.0. 

window : str or tuple or array_like, optional 

Desired window to use. If `window` is a string or tuple, it is 

passed to `get_window` to generate the window values, which are 

DFT-even by default. See `get_window` for a list of windows and 

required parameters. If `window` is array_like it will be used 

directly as the window and its length must be nperseg. Defaults 

to a Hann window. 

nperseg : int, optional 

Length of each segment. Defaults to 256. 

noverlap : int, optional 

Number of points to overlap between segments. If `None`, 

``noverlap = nperseg // 2``. Defaults to `None`. When 

specified, the COLA constraint must be met (see Notes below). 

nfft : int, optional 

Length of the FFT used, if a zero padded FFT is desired. If 

`None`, the FFT length is `nperseg`. Defaults to `None`. 

detrend : str or function or `False`, optional 

Specifies how to detrend each segment. If `detrend` is a 

string, it is passed as the `type` argument to the `detrend` 

function. If it is a function, it takes a segment and returns a 

detrended segment. If `detrend` is `False`, no detrending is 

done. Defaults to `False`. 

return_onesided : bool, optional 

If `True`, return a one-sided spectrum for real data. If 

`False` return a two-sided spectrum. Note that for complex 

data, a two-sided spectrum is always returned. Defaults to 

`True`. 

boundary : str or None, optional 

Specifies whether the input signal is extended at both ends, and 

how to generate the new values, in order to center the first 

windowed segment on the first input point. This has the benefit 

of enabling reconstruction of the first input point when the 

employed window function starts at zero. Valid options are 

``['even', 'odd', 'constant', 'zeros', None]``. Defaults to 

'zeros', for zero padding extension. I.e. ``[1, 2, 3, 4]`` is 

extended to ``[0, 1, 2, 3, 4, 0]`` for ``nperseg=3``. 

padded : bool, optional 

Specifies whether the input signal is zero-padded at the end to 

make the signal fit exactly into an integer number of window 

segments, so that all of the signal is included in the output. 

Defaults to `True`. Padding occurs after boundary extension, if 

`boundary` is not `None`, and `padded` is `True`, as is the 

default. 

axis : int, optional 

Axis along which the STFT is computed; the default is over the 

last axis (i.e. ``axis=-1``). 

 

Returns 

------- 

f : ndarray 

Array of sample frequencies. 

t : ndarray 

Array of segment times. 

Zxx : ndarray 

STFT of `x`. By default, the last axis of `Zxx` corresponds 

to the segment times. 

 

See Also 

-------- 

istft: Inverse Short Time Fourier Transform 

check_COLA: Check whether the Constant OverLap Add (COLA) constraint 

is met 

welch: Power spectral density by Welch's method. 

spectrogram: Spectrogram by Welch's method. 

csd: Cross spectral density by Welch's method. 

lombscargle: Lomb-Scargle periodogram for unevenly sampled data 

 

Notes 

----- 

In order to enable inversion of an STFT via the inverse STFT in 

`istft`, the signal windowing must obey the constraint of "Constant 

OverLap Add" (COLA), and the input signal must have complete 

windowing coverage (i.e. ``(x.shape[axis] - nperseg) % 

(nperseg-noverlap) == 0``). The `padded` argument may be used to 

accomplish this. 

 

The COLA constraint ensures that every point in the input data is 

equally weighted, thereby avoiding aliasing and allowing full 

reconstruction. Whether a choice of `window`, `nperseg`, and 

`noverlap` satisfy this constraint can be tested with 

`check_COLA`. 

 

.. versionadded:: 0.19.0 

 

References 

---------- 

.. [1] Oppenheim, Alan V., Ronald W. Schafer, John R. Buck 

"Discrete-Time Signal Processing", Prentice Hall, 1999. 

.. [2] Daniel W. Griffin, Jae S. Limdt "Signal Estimation from 

Modified Short Fourier Transform", IEEE 1984, 

10.1109/TASSP.1984.1164317 

 

Examples 

-------- 

>>> from scipy import signal 

>>> import matplotlib.pyplot as plt 

 

Generate a test signal, a 2 Vrms sine wave whose frequency is slowly 

modulated around 3kHz, corrupted by white noise of exponentially 

decreasing magnitude sampled at 10 kHz. 

 

>>> fs = 10e3 

>>> N = 1e5 

>>> amp = 2 * np.sqrt(2) 

>>> noise_power = 0.01 * fs / 2 

>>> time = np.arange(N) / float(fs) 

>>> mod = 500*np.cos(2*np.pi*0.25*time) 

>>> carrier = amp * np.sin(2*np.pi*3e3*time + mod) 

>>> noise = np.random.normal(scale=np.sqrt(noise_power), 

... size=time.shape) 

>>> noise *= np.exp(-time/5) 

>>> x = carrier + noise 

 

Compute and plot the STFT's magnitude. 

 

>>> f, t, Zxx = signal.stft(x, fs, nperseg=1000) 

>>> plt.pcolormesh(t, f, np.abs(Zxx), vmin=0, vmax=amp) 

>>> plt.title('STFT Magnitude') 

>>> plt.ylabel('Frequency [Hz]') 

>>> plt.xlabel('Time [sec]') 

>>> plt.show() 

""" 

 

freqs, time, Zxx = _spectral_helper(x, x, fs, window, nperseg, noverlap, 

nfft, detrend, return_onesided, 

scaling='spectrum', axis=axis, 

mode='stft', boundary=boundary, 

padded=padded) 

 

return freqs, time, Zxx 

 

 

def istft(Zxx, fs=1.0, window='hann', nperseg=None, noverlap=None, nfft=None, 

input_onesided=True, boundary=True, time_axis=-1, freq_axis=-2): 

r""" 

Perform the inverse Short Time Fourier transform (iSTFT). 

 

Parameters 

---------- 

Zxx : array_like 

STFT of the signal to be reconstructed. If a purely real array 

is passed, it will be cast to a complex data type. 

fs : float, optional 

Sampling frequency of the time series. Defaults to 1.0. 

window : str or tuple or array_like, optional 

Desired window to use. If `window` is a string or tuple, it is 

passed to `get_window` to generate the window values, which are 

DFT-even by default. See `get_window` for a list of windows and 

required parameters. If `window` is array_like it will be used 

directly as the window and its length must be nperseg. Defaults 

to a Hann window. Must match the window used to generate the 

STFT for faithful inversion. 

nperseg : int, optional 

Number of data points corresponding to each STFT segment. This 

parameter must be specified if the number of data points per 

segment is odd, or if the STFT was padded via ``nfft > 

nperseg``. If `None`, the value depends on the shape of 

`Zxx` and `input_onesided`. If `input_onesided` is True, 

``nperseg=2*(Zxx.shape[freq_axis] - 1)``. Otherwise, 

``nperseg=Zxx.shape[freq_axis]``. Defaults to `None`. 

noverlap : int, optional 

Number of points to overlap between segments. If `None`, half 

of the segment length. Defaults to `None`. When specified, the 

COLA constraint must be met (see Notes below), and should match 

the parameter used to generate the STFT. Defaults to `None`. 

nfft : int, optional 

Number of FFT points corresponding to each STFT segment. This 

parameter must be specified if the STFT was padded via ``nfft > 

nperseg``. If `None`, the default values are the same as for 

`nperseg`, detailed above, with one exception: if 

`input_onesided` is True and 

``nperseg==2*Zxx.shape[freq_axis] - 1``, `nfft` also takes on 

that value. This case allows the proper inversion of an 

odd-length unpadded STFT using ``nfft=None``. Defaults to 

`None`. 

input_onesided : bool, optional 

If `True`, interpret the input array as one-sided FFTs, such 

as is returned by `stft` with ``return_onesided=True`` and 

`numpy.fft.rfft`. If `False`, interpret the input as a a 

two-sided FFT. Defaults to `True`. 

boundary : bool, optional 

Specifies whether the input signal was extended at its 

boundaries by supplying a non-`None` ``boundary`` argument to 

`stft`. Defaults to `True`. 

time_axis : int, optional 

Where the time segments of the STFT is located; the default is 

the last axis (i.e. ``axis=-1``). 

freq_axis : int, optional 

Where the frequency axis of the STFT is located; the default is 

the penultimate axis (i.e. ``axis=-2``). 

 

Returns 

------- 

t : ndarray 

Array of output data times. 

x : ndarray 

iSTFT of `Zxx`. 

 

See Also 

-------- 

stft: Short Time Fourier Transform 

check_COLA: Check whether the Constant OverLap Add (COLA) constraint 

is met 

 

Notes 

----- 

In order to enable inversion of an STFT via the inverse STFT with 

`istft`, the signal windowing must obey the constraint of "Constant 

OverLap Add" (COLA). This ensures that every point in the input data 

is equally weighted, thereby avoiding aliasing and allowing full 

reconstruction. Whether a choice of `window`, `nperseg`, and 

`noverlap` satisfy this constraint can be tested with 

`check_COLA`, by using ``nperseg = Zxx.shape[freq_axis]``. 

 

An STFT which has been modified (via masking or otherwise) is not 

guaranteed to correspond to a exactly realizible signal. This 

function implements the iSTFT via the least-squares esimation 

algorithm detailed in [2]_, which produces a signal that minimizes 

the mean squared error between the STFT of the returned signal and 

the modified STFT. 

 

.. versionadded:: 0.19.0 

 

References 

---------- 

.. [1] Oppenheim, Alan V., Ronald W. Schafer, John R. Buck 

"Discrete-Time Signal Processing", Prentice Hall, 1999. 

.. [2] Daniel W. Griffin, Jae S. Limdt "Signal Estimation from 

Modified Short Fourier Transform", IEEE 1984, 

10.1109/TASSP.1984.1164317 

 

Examples 

-------- 

>>> from scipy import signal 

>>> import matplotlib.pyplot as plt 

 

Generate a test signal, a 2 Vrms sine wave at 50Hz corrupted by 

0.001 V**2/Hz of white noise sampled at 1024 Hz. 

 

>>> fs = 1024 

>>> N = 10*fs 

>>> nperseg = 512 

>>> amp = 2 * np.sqrt(2) 

>>> noise_power = 0.001 * fs / 2 

>>> time = np.arange(N) / float(fs) 

>>> carrier = amp * np.sin(2*np.pi*50*time) 

>>> noise = np.random.normal(scale=np.sqrt(noise_power), 

... size=time.shape) 

>>> x = carrier + noise 

 

Compute the STFT, and plot its magnitude 

 

>>> f, t, Zxx = signal.stft(x, fs=fs, nperseg=nperseg) 

>>> plt.figure() 

>>> plt.pcolormesh(t, f, np.abs(Zxx), vmin=0, vmax=amp) 

>>> plt.ylim([f[1], f[-1]]) 

>>> plt.title('STFT Magnitude') 

>>> plt.ylabel('Frequency [Hz]') 

>>> plt.xlabel('Time [sec]') 

>>> plt.yscale('log') 

>>> plt.show() 

 

Zero the components that are 10% or less of the carrier magnitude, 

then convert back to a time series via inverse STFT 

 

>>> Zxx = np.where(np.abs(Zxx) >= amp/10, Zxx, 0) 

>>> _, xrec = signal.istft(Zxx, fs) 

 

Compare the cleaned signal with the original and true carrier signals. 

 

>>> plt.figure() 

>>> plt.plot(time, x, time, xrec, time, carrier) 

>>> plt.xlim([2, 2.1]) 

>>> plt.xlabel('Time [sec]') 

>>> plt.ylabel('Signal') 

>>> plt.legend(['Carrier + Noise', 'Filtered via STFT', 'True Carrier']) 

>>> plt.show() 

 

Note that the cleaned signal does not start as abruptly as the original, 

since some of the coefficients of the transient were also removed: 

 

>>> plt.figure() 

>>> plt.plot(time, x, time, xrec, time, carrier) 

>>> plt.xlim([0, 0.1]) 

>>> plt.xlabel('Time [sec]') 

>>> plt.ylabel('Signal') 

>>> plt.legend(['Carrier + Noise', 'Filtered via STFT', 'True Carrier']) 

>>> plt.show() 

 

""" 

 

# Make sure input is an ndarray of appropriate complex dtype 

Zxx = np.asarray(Zxx) + 0j 

freq_axis = int(freq_axis) 

time_axis = int(time_axis) 

 

if Zxx.ndim < 2: 

raise ValueError('Input stft must be at least 2d!') 

 

if freq_axis == time_axis: 

raise ValueError('Must specify differing time and frequency axes!') 

 

nseg = Zxx.shape[time_axis] 

 

if input_onesided: 

# Assume even segment length 

n_default = 2*(Zxx.shape[freq_axis] - 1) 

else: 

n_default = Zxx.shape[freq_axis] 

 

# Check windowing parameters 

if nperseg is None: 

nperseg = n_default 

else: 

nperseg = int(nperseg) 

if nperseg < 1: 

raise ValueError('nperseg must be a positive integer') 

 

if nfft is None: 

if (input_onesided) and (nperseg == n_default + 1): 

# Odd nperseg, no FFT padding 

nfft = nperseg 

else: 

nfft = n_default 

elif nfft < nperseg: 

raise ValueError('nfft must be greater than or equal to nperseg.') 

else: 

nfft = int(nfft) 

 

if noverlap is None: 

noverlap = nperseg//2 

else: 

noverlap = int(noverlap) 

if noverlap >= nperseg: 

raise ValueError('noverlap must be less than nperseg.') 

nstep = nperseg - noverlap 

 

if not check_COLA(window, nperseg, noverlap): 

raise ValueError('Window, STFT shape and noverlap do not satisfy the ' 

'COLA constraint.') 

 

# Rearrange axes if necessary 

if time_axis != Zxx.ndim-1 or freq_axis != Zxx.ndim-2: 

# Turn negative indices to positive for the call to transpose 

if freq_axis < 0: 

freq_axis = Zxx.ndim + freq_axis 

if time_axis < 0: 

time_axis = Zxx.ndim + time_axis 

zouter = list(range(Zxx.ndim)) 

for ax in sorted([time_axis, freq_axis], reverse=True): 

zouter.pop(ax) 

Zxx = np.transpose(Zxx, zouter+[freq_axis, time_axis]) 

 

# Get window as array 

if isinstance(window, string_types) or type(window) is tuple: 

win = get_window(window, nperseg) 

else: 

win = np.asarray(window) 

if len(win.shape) != 1: 

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

if win.shape[0] != nperseg: 

raise ValueError('window must have length of {0}'.format(nperseg)) 

 

if input_onesided: 

ifunc = np.fft.irfft 

else: 

ifunc = fftpack.ifft 

 

xsubs = ifunc(Zxx, axis=-2, n=nfft)[..., :nperseg, :] 

 

# Initialize output and normalization arrays 

outputlength = nperseg + (nseg-1)*nstep 

x = np.zeros(list(Zxx.shape[:-2])+[outputlength], dtype=xsubs.dtype) 

norm = np.zeros(outputlength, dtype=xsubs.dtype) 

 

if np.result_type(win, xsubs) != xsubs.dtype: 

win = win.astype(xsubs.dtype) 

 

xsubs *= win.sum() # This takes care of the 'spectrum' scaling 

 

# Construct the output from the ifft segments 

# This loop could perhaps be vectorized/strided somehow... 

for ii in range(nseg): 

# Window the ifft 

x[..., ii*nstep:ii*nstep+nperseg] += xsubs[..., ii] * win 

norm[..., ii*nstep:ii*nstep+nperseg] += win**2 

 

# Divide out normalization where non-tiny 

x /= np.where(norm > 1e-10, norm, 1.0) 

 

# Remove extension points 

if boundary: 

x = x[..., nperseg//2:-(nperseg//2)] 

 

if input_onesided: 

x = x.real 

 

# Put axes back 

if x.ndim > 1: 

if time_axis != Zxx.ndim-1: 

if freq_axis < time_axis: 

time_axis -= 1 

x = np.rollaxis(x, -1, time_axis) 

 

time = np.arange(x.shape[0])/float(fs) 

return time, x 

 

 

def coherence(x, y, fs=1.0, window='hann', nperseg=None, noverlap=None, 

nfft=None, detrend='constant', axis=-1): 

r""" 

Estimate the magnitude squared coherence estimate, Cxy, of 

discrete-time signals X and Y using Welch's method. 

 

``Cxy = abs(Pxy)**2/(Pxx*Pyy)``, where `Pxx` and `Pyy` are power 

spectral density estimates of X and Y, and `Pxy` is the cross 

spectral density estimate of X and Y. 

 

Parameters 

---------- 

x : array_like 

Time series of measurement values 

y : array_like 

Time series of measurement values 

fs : float, optional 

Sampling frequency of the `x` and `y` time series. Defaults 

to 1.0. 

window : str or tuple or array_like, optional 

Desired window to use. If `window` is a string or tuple, it is 

passed to `get_window` to generate the window values, which are 

DFT-even by default. See `get_window` for a list of windows and 

required parameters. If `window` is array_like it will be used 

directly as the window and its length must be nperseg. Defaults 

to a Hann window. 

nperseg : int, optional 

Length of each segment. Defaults to None, but if window is str or 

tuple, is set to 256, and if window is array_like, is set to the 

length of the window. 

noverlap: int, optional 

Number of points to overlap between segments. If `None`, 

``noverlap = nperseg // 2``. Defaults to `None`. 

nfft : int, optional 

Length of the FFT used, if a zero padded FFT is desired. If 

`None`, the FFT length is `nperseg`. Defaults to `None`. 

detrend : str or function or `False`, optional 

Specifies how to detrend each segment. If `detrend` is a 

string, it is passed as the `type` argument to the `detrend` 

function. If it is a function, it takes a segment and returns a 

detrended segment. If `detrend` is `False`, no detrending is 

done. Defaults to 'constant'. 

axis : int, optional 

Axis along which the coherence is computed for both inputs; the 

default is over the last axis (i.e. ``axis=-1``). 

 

Returns 

------- 

f : ndarray 

Array of sample frequencies. 

Cxy : ndarray 

Magnitude squared coherence of x and y. 

 

See Also 

-------- 

periodogram: Simple, optionally modified periodogram 

lombscargle: Lomb-Scargle periodogram for unevenly sampled data 

welch: Power spectral density by Welch's method. 

csd: Cross spectral density by Welch's method. 

 

Notes 

-------- 

An appropriate amount of overlap will depend on the choice of window 

and on your requirements. For the default Hann window an overlap of 

50% is a reasonable trade off between accurately estimating the 

signal power, while not over counting any of the data. Narrower 

windows may require a larger overlap. 

 

.. versionadded:: 0.16.0 

 

References 

---------- 

.. [1] P. Welch, "The use of the fast Fourier transform for the 

estimation of power spectra: A method based on time averaging 

over short, modified periodograms", IEEE Trans. Audio 

Electroacoust. vol. 15, pp. 70-73, 1967. 

.. [2] Stoica, Petre, and Randolph Moses, "Spectral Analysis of 

Signals" Prentice Hall, 2005 

 

Examples 

-------- 

>>> from scipy import signal 

>>> import matplotlib.pyplot as plt 

 

Generate two test signals with some common features. 

 

>>> fs = 10e3 

>>> N = 1e5 

>>> amp = 20 

>>> freq = 1234.0 

>>> noise_power = 0.001 * fs / 2 

>>> time = np.arange(N) / fs 

>>> b, a = signal.butter(2, 0.25, 'low') 

>>> x = np.random.normal(scale=np.sqrt(noise_power), size=time.shape) 

>>> y = signal.lfilter(b, a, x) 

>>> x += amp*np.sin(2*np.pi*freq*time) 

>>> y += np.random.normal(scale=0.1*np.sqrt(noise_power), size=time.shape) 

 

Compute and plot the coherence. 

 

>>> f, Cxy = signal.coherence(x, y, fs, nperseg=1024) 

>>> plt.semilogy(f, Cxy) 

>>> plt.xlabel('frequency [Hz]') 

>>> plt.ylabel('Coherence') 

>>> plt.show() 

""" 

 

freqs, Pxx = welch(x, fs, window, nperseg, noverlap, nfft, detrend, 

axis=axis) 

_, Pyy = welch(y, fs, window, nperseg, noverlap, nfft, detrend, axis=axis) 

_, Pxy = csd(x, y, fs, window, nperseg, noverlap, nfft, detrend, axis=axis) 

 

Cxy = np.abs(Pxy)**2 / Pxx / Pyy 

 

return freqs, Cxy 

 

 

def _spectral_helper(x, y, fs=1.0, window='hann', nperseg=None, noverlap=None, 

nfft=None, detrend='constant', return_onesided=True, 

scaling='spectrum', axis=-1, mode='psd', boundary=None, 

padded=False): 

""" 

Calculate various forms of windowed FFTs for PSD, CSD, etc. 

 

This is a helper function that implements the commonality between 

the stft, psd, csd, and spectrogram functions. It is not designed to 

be called externally. The windows are not averaged over; the result 

from each window is returned. 

 

Parameters 

--------- 

x : array_like 

Array or sequence containing the data to be analyzed. 

y : array_like 

Array or sequence containing the data to be analyzed. If this is 

the same object in memory as `x` (i.e. ``_spectral_helper(x, 

x, ...)``), the extra computations are spared. 

fs : float, optional 

Sampling frequency of the time series. Defaults to 1.0. 

window : str or tuple or array_like, optional 

Desired window to use. If `window` is a string or tuple, it is 

passed to `get_window` to generate the window values, which are 

DFT-even by default. See `get_window` for a list of windows and 

required parameters. If `window` is array_like it will be used 

directly as the window and its length must be nperseg. Defaults 

to a Hann window. 

nperseg : int, optional 

Length of each segment. Defaults to None, but if window is str or 

tuple, is set to 256, and if window is array_like, is set to the 

length of the window. 

noverlap : int, optional 

Number of points to overlap between segments. If `None`, 

``noverlap = nperseg // 2``. Defaults to `None`. 

nfft : int, optional 

Length of the FFT used, if a zero padded FFT is desired. If 

`None`, the FFT length is `nperseg`. Defaults to `None`. 

detrend : str or function or `False`, optional 

Specifies how to detrend each segment. If `detrend` is a 

string, it is passed as the `type` argument to the `detrend` 

function. If it is a function, it takes a segment and returns a 

detrended segment. If `detrend` is `False`, no detrending is 

done. Defaults to 'constant'. 

return_onesided : bool, optional 

If `True`, return a one-sided spectrum for real data. If 

`False` return a two-sided spectrum. Note that for complex 

data, a two-sided spectrum is always returned. 

scaling : { 'density', 'spectrum' }, optional 

Selects between computing the cross spectral density ('density') 

where `Pxy` has units of V**2/Hz and computing the cross 

spectrum ('spectrum') where `Pxy` has units of V**2, if `x` 

and `y` are measured in V and `fs` is measured in Hz. 

Defaults to 'density' 

axis : int, optional 

Axis along which the FFTs are computed; the default is over the 

last axis (i.e. ``axis=-1``). 

mode: str {'psd', 'stft'}, optional 

Defines what kind of return values are expected. Defaults to 

'psd'. 

boundary : str or None, optional 

Specifies whether the input signal is extended at both ends, and 

how to generate the new values, in order to center the first 

windowed segment on the first input point. This has the benefit 

of enabling reconstruction of the first input point when the 

employed window function starts at zero. Valid options are 

``['even', 'odd', 'constant', 'zeros', None]``. Defaults to 

`None`. 

padded : bool, optional 

Specifies whether the input signal is zero-padded at the end to 

make the signal fit exactly into an integer number of window 

segments, so that all of the signal is included in the output. 

Defaults to `False`. Padding occurs after boundary extension, if 

`boundary` is not `None`, and `padded` is `True`. 

Returns 

------- 

freqs : ndarray 

Array of sample frequencies. 

t : ndarray 

Array of times corresponding to each data segment 

result : ndarray 

Array of output data, contents dependent on *mode* kwarg. 

 

Notes 

----- 

Adapted from matplotlib.mlab 

 

.. versionadded:: 0.16.0 

""" 

if mode not in ['psd', 'stft']: 

raise ValueError("Unknown value for mode %s, must be one of: " 

"{'psd', 'stft'}" % mode) 

 

boundary_funcs = {'even': even_ext, 

'odd': odd_ext, 

'constant': const_ext, 

'zeros': zero_ext, 

None: None} 

 

if boundary not in boundary_funcs: 

raise ValueError("Unknown boundary option '{0}', must be one of: {1}" 

.format(boundary, list(boundary_funcs.keys()))) 

 

# If x and y are the same object we can save ourselves some computation. 

same_data = y is x 

 

if not same_data and mode != 'psd': 

raise ValueError("x and y must be equal if mode is 'stft'") 

 

axis = int(axis) 

 

# Ensure we have np.arrays, get outdtype 

x = np.asarray(x) 

if not same_data: 

y = np.asarray(y) 

outdtype = np.result_type(x, y, np.complex64) 

else: 

outdtype = np.result_type(x, np.complex64) 

 

if not same_data: 

# Check if we can broadcast the outer axes together 

xouter = list(x.shape) 

youter = list(y.shape) 

xouter.pop(axis) 

youter.pop(axis) 

try: 

outershape = np.broadcast(np.empty(xouter), np.empty(youter)).shape 

except ValueError: 

raise ValueError('x and y cannot be broadcast together.') 

 

if same_data: 

if x.size == 0: 

return np.empty(x.shape), np.empty(x.shape), np.empty(x.shape) 

else: 

if x.size == 0 or y.size == 0: 

outshape = outershape + (min([x.shape[axis], y.shape[axis]]),) 

emptyout = np.rollaxis(np.empty(outshape), -1, axis) 

return emptyout, emptyout, emptyout 

 

if x.ndim > 1: 

if axis != -1: 

x = np.rollaxis(x, axis, len(x.shape)) 

if not same_data and y.ndim > 1: 

y = np.rollaxis(y, axis, len(y.shape)) 

 

# Check if x and y are the same length, zero-pad if necessary 

if not same_data: 

if x.shape[-1] != y.shape[-1]: 

if x.shape[-1] < y.shape[-1]: 

pad_shape = list(x.shape) 

pad_shape[-1] = y.shape[-1] - x.shape[-1] 

x = np.concatenate((x, np.zeros(pad_shape)), -1) 

else: 

pad_shape = list(y.shape) 

pad_shape[-1] = x.shape[-1] - y.shape[-1] 

y = np.concatenate((y, np.zeros(pad_shape)), -1) 

 

if nperseg is not None: # if specified by user 

nperseg = int(nperseg) 

if nperseg < 1: 

raise ValueError('nperseg must be a positive integer') 

 

# parse window; if array like, then set nperseg = win.shape 

win, nperseg = _triage_segments(window, nperseg,input_length=x.shape[-1]) 

 

if nfft is None: 

nfft = nperseg 

elif nfft < nperseg: 

raise ValueError('nfft must be greater than or equal to nperseg.') 

else: 

nfft = int(nfft) 

 

if noverlap is None: 

noverlap = nperseg//2 

else: 

noverlap = int(noverlap) 

if noverlap >= nperseg: 

raise ValueError('noverlap must be less than nperseg.') 

nstep = nperseg - noverlap 

 

# Padding occurs after boundary extension, so that the extended signal ends 

# in zeros, instead of introducing an impulse at the end. 

# I.e. if x = [..., 3, 2] 

# extend then pad -> [..., 3, 2, 2, 3, 0, 0, 0] 

# pad then extend -> [..., 3, 2, 0, 0, 0, 2, 3] 

 

if boundary is not None: 

ext_func = boundary_funcs[boundary] 

x = ext_func(x, nperseg//2, axis=-1) 

if not same_data: 

y = ext_func(y, nperseg//2, axis=-1) 

 

if padded: 

# Pad to integer number of windowed segments 

# I.e make x.shape[-1] = nperseg + (nseg-1)*nstep, with integer nseg 

nadd = (-(x.shape[-1]-nperseg) % nstep) % nperseg 

zeros_shape = list(x.shape[:-1]) + [nadd] 

x = np.concatenate((x, np.zeros(zeros_shape)), axis=-1) 

if not same_data: 

zeros_shape = list(y.shape[:-1]) + [nadd] 

y = np.concatenate((y, np.zeros(zeros_shape)), axis=-1) 

 

# Handle detrending and window functions 

if not detrend: 

def detrend_func(d): 

return d 

elif not hasattr(detrend, '__call__'): 

def detrend_func(d): 

return signaltools.detrend(d, type=detrend, axis=-1) 

elif axis != -1: 

# Wrap this function so that it receives a shape that it could 

# reasonably expect to receive. 

def detrend_func(d): 

d = np.rollaxis(d, -1, axis) 

d = detrend(d) 

return np.rollaxis(d, axis, len(d.shape)) 

else: 

detrend_func = detrend 

 

if np.result_type(win,np.complex64) != outdtype: 

win = win.astype(outdtype) 

 

if scaling == 'density': 

scale = 1.0 / (fs * (win*win).sum()) 

elif scaling == 'spectrum': 

scale = 1.0 / win.sum()**2 

else: 

raise ValueError('Unknown scaling: %r' % scaling) 

 

if mode == 'stft': 

scale = np.sqrt(scale) 

 

if return_onesided: 

if np.iscomplexobj(x): 

sides = 'twosided' 

warnings.warn('Input data is complex, switching to ' 

'return_onesided=False') 

else: 

sides = 'onesided' 

if not same_data: 

if np.iscomplexobj(y): 

sides = 'twosided' 

warnings.warn('Input data is complex, switching to ' 

'return_onesided=False') 

else: 

sides = 'twosided' 

 

if sides == 'twosided': 

freqs = fftpack.fftfreq(nfft, 1/fs) 

elif sides == 'onesided': 

freqs = np.fft.rfftfreq(nfft, 1/fs) 

 

# Perform the windowed FFTs 

result = _fft_helper(x, win, detrend_func, nperseg, noverlap, nfft, sides) 

 

if not same_data: 

# All the same operations on the y data 

result_y = _fft_helper(y, win, detrend_func, nperseg, noverlap, nfft, 

sides) 

result = np.conjugate(result) * result_y 

elif mode == 'psd': 

result = np.conjugate(result) * result 

 

result *= scale 

if sides == 'onesided' and mode == 'psd': 

if nfft % 2: 

result[..., 1:] *= 2 

else: 

# Last point is unpaired Nyquist freq point, don't double 

result[..., 1:-1] *= 2 

 

time = np.arange(nperseg/2, x.shape[-1] - nperseg/2 + 1, 

nperseg - noverlap)/float(fs) 

if boundary is not None: 

time -= (nperseg/2) / fs 

 

result = result.astype(outdtype) 

 

# All imaginary parts are zero anyways 

if same_data and mode != 'stft': 

result = result.real 

 

# Output is going to have new last axis for time/window index, so a 

# negative axis index shifts down one 

if axis < 0: 

axis -= 1 

 

# Roll frequency axis back to axis where the data came from 

result = np.rollaxis(result, -1, axis) 

 

return freqs, time, result 

 

 

def _fft_helper(x, win, detrend_func, nperseg, noverlap, nfft, sides): 

""" 

Calculate windowed FFT, for internal use by 

scipy.signal._spectral_helper 

 

This is a helper function that does the main FFT calculation for 

`_spectral helper`. All input validation is performed there, and the 

data axis is assumed to be the last axis of x. It is not designed to 

be called externally. The windows are not averaged over; the result 

from each window is returned. 

 

Returns 

------- 

result : ndarray 

Array of FFT data 

 

Notes 

----- 

Adapted from matplotlib.mlab 

 

.. versionadded:: 0.16.0 

""" 

# Created strided array of data segments 

if nperseg == 1 and noverlap == 0: 

result = x[..., np.newaxis] 

else: 

# http://stackoverflow.com/a/5568169 

step = nperseg - noverlap 

shape = x.shape[:-1]+((x.shape[-1]-noverlap)//step, nperseg) 

strides = x.strides[:-1]+(step*x.strides[-1], x.strides[-1]) 

result = np.lib.stride_tricks.as_strided(x, shape=shape, 

strides=strides) 

 

# Detrend each data segment individually 

result = detrend_func(result) 

 

# Apply window by multiplication 

result = win * result 

 

# Perform the fft. Acts on last axis by default. Zero-pads automatically 

if sides == 'twosided': 

func = fftpack.fft 

else: 

result = result.real 

func = np.fft.rfft 

result = func(result, n=nfft) 

 

return result 

 

def _triage_segments(window, nperseg,input_length): 

""" 

Parses window and nperseg arguments for spectrogram and _spectral_helper. 

This is a helper function, not meant to be called externally. 

 

Parameters 

--------- 

window : string, tuple, or ndarray 

If window is specified by a string or tuple and nperseg is not 

specified, nperseg is set to the default of 256 and returns a window of 

that length. 

If instead the window is array_like and nperseg is not specified, then 

nperseg is set to the length of the window. A ValueError is raised if 

the user supplies both an array_like window and a value for nperseg but 

nperseg does not equal the length of the window. 

 

nperseg : int 

Length of each segment 

 

input_length: int 

Length of input signal, i.e. x.shape[-1]. Used to test for errors. 

 

Returns 

------- 

win : ndarray 

window. If function was called with string or tuple than this will hold 

the actual array used as a window. 

 

nperseg : int 

Length of each segment. If window is str or tuple, nperseg is set to 

256. If window is array_like, nperseg is set to the length of the 

6 

window. 

""" 

 

#parse window; if array like, then set nperseg = win.shape 

if isinstance(window, string_types) or isinstance(window, tuple): 

# if nperseg not specified 

if nperseg is None: 

nperseg = 256 # then change to default 

if nperseg > input_length: 

warnings.warn('nperseg = {0:d} is greater than input length ' 

' = {1:d}, using nperseg = {1:d}' 

.format(nperseg, input_length)) 

nperseg = input_length 

win = get_window(window, nperseg) 

else: 

win = np.asarray(window) 

if len(win.shape) != 1: 

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

if input_length < win.shape[-1]: 

raise ValueError('window is longer than input signal') 

if nperseg is None: 

nperseg = win.shape[0] 

elif nperseg is not None: 

if nperseg != win.shape[0]: 

raise ValueError("value specified for nperseg is different from" 

" length of window") 

return win, nperseg