File size: 14,705 Bytes
7885a28
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# pylint: disable=missing-docstring
import numpy as np
import pytest
from scipy._lib._array_api import xp_assert_close

from scipy.signal._spline import (
    symiirorder1_ic, symiirorder2_ic_fwd, symiirorder2_ic_bwd)
from scipy.signal import symiirorder1, symiirorder2


def _compute_symiirorder2_bwd_hs(k, cs, rsq, omega):
    cssq = cs * cs
    k = np.abs(k)
    rsupk = np.power(rsq, k / 2.0)

    c0 = (cssq * (1.0 + rsq) / (1.0 - rsq) /
          (1 - 2 * rsq * np.cos(2 * omega) + rsq * rsq))
    gamma = (1.0 - rsq) / (1.0 + rsq) / np.tan(omega)
    return c0 * rsupk * (np.cos(omega * k) + gamma * np.sin(omega * k))


class TestSymIIR:
    @pytest.mark.parametrize(
        'dtype', [np.float32, np.float64, np.complex64, np.complex128])
    @pytest.mark.parametrize('precision', [-1.0, 0.7, 0.5, 0.25, 0.0075])
    def test_symiir1_ic(self, dtype, precision):
        c_precision = precision
        if precision <= 0.0 or precision > 1.0:
            if dtype in {np.float32, np.complex64}:
                c_precision = 1e-6
            else:
                c_precision = 1e-11

        # Symmetrical initial conditions for a IIR filter of order 1 are:
        # x[0] + z1 * \sum{k = 0}^{n - 1} x[k] * z1^k

        # Check the initial condition for a low-pass filter
        # with coefficient b = 0.85 on a step signal. The initial condition is
        # a geometric series: 1 + b * \sum_{k = 0}^{n - 1} u[k] b^k.

        # Finding the initial condition corresponds to
        # 1. Computing the index n such that b**n < precision, which
        # corresponds to ceil(log(precision) / log(b))
        # 2. Computing the geometric series until n, this can be computed
        # using the partial sum formula: (1 - b**n) / (1 - b)
        # This holds due to the input being a step signal.
        b = 0.85
        n_exp = int(np.ceil(np.log(c_precision) / np.log(b)))
        expected = np.asarray([[(1 - b ** n_exp) / (1 - b)]], dtype=dtype)
        expected = 1 + b * expected

        # Create a step signal of size n + 1
        x = np.ones(n_exp + 1, dtype=dtype)
        xp_assert_close(symiirorder1_ic(x, b, precision), expected,
                        atol=2e-6, rtol=2e-7)

        # Check the conditions for a exponential decreasing signal with base 2.
        # Same conditions hold, as the product of 0.5^n * 0.85^n is
        # still a geometric series
        b_d = np.asarray(b, dtype=dtype)
        expected = np.asarray(
            [[(1 - (0.5 * b_d) ** n_exp) / (1 - (0.5 * b_d))]], dtype=dtype)
        expected = 1 + b_d * expected

        # Create an exponential decreasing signal of size n + 1
        x = 2 ** -np.arange(n_exp + 1, dtype=dtype)
        xp_assert_close(symiirorder1_ic(x, b, precision), expected,
                        atol=2e-6, rtol=2e-7)

    def test_symiir1_ic_fails(self):
        # Test that symiirorder1_ic fails whenever \sum_{n = 1}^{n} b^n > eps
        b = 0.85
        # Create a step signal of size 100
        x = np.ones(100, dtype=np.float64)

        # Compute the closed form for the geometrical series
        precision = 1 / (1 - b)
        pytest.raises(ValueError, symiirorder1_ic, x, b, precision)

        # Test that symiirorder1_ic fails when |z1| >= 1
        pytest.raises(ValueError, symiirorder1_ic, x, 1.0, -1)
        pytest.raises(ValueError, symiirorder1_ic, x, 2.0, -1)

    @pytest.mark.parametrize(
        'dtype', [np.float32, np.float64, np.complex64, np.complex128])
    @pytest.mark.parametrize('precision', [-1.0, 0.7, 0.5, 0.25, 0.0075])
    def test_symiir1(self, dtype, precision):
        c_precision = precision
        if precision <= 0.0 or precision > 1.0:
            if dtype in {np.float32, np.complex64}:
                c_precision = 1e-6
            else:
                c_precision = 1e-11

        # Test for a low-pass filter with c0 = 0.15 and z1 = 0.85
        # using an unit step over 200 samples.
        c0 = 0.15
        z1 = 0.85
        n = 200
        signal = np.ones(n, dtype=dtype)

        # Find the initial condition. See test_symiir1_ic for a detailed
        # explanation
        n_exp = int(np.ceil(np.log(c_precision) / np.log(z1)))
        initial = np.asarray((1 - z1 ** n_exp) / (1 - z1), dtype=dtype)
        initial = 1 + z1 * initial

        # Forward pass
        # The transfer function for the system 1 / (1 - z1 * z^-1) when
        # applied to an unit step with initial conditions y0 is
        # 1 / (1 - z1 * z^-1) * (z^-1 / (1 - z^-1) + y0)

        # Solving the inverse Z-transform for the given expression yields:
        # y[n] = y0 * z1**n * u[n] +
        #        -z1 / (1 - z1) * z1**(k - 1) * u[k - 1] +
        #        1 / (1 - z1) * u[k - 1]
        # d is the Kronecker delta function, and u is the unit step

        # y0 * z1**n * u[n]
        pos = np.arange(n, dtype=dtype)
        comp1 = initial * z1**pos

        # -z1 / (1 - z1) * z1**(k - 1) * u[k - 1]
        comp2 = np.zeros(n, dtype=dtype)
        comp2[1:] = -z1 / (1 - z1) * z1**pos[:-1]

        # 1 / (1 - z1) * u[k - 1]
        comp3 = np.zeros(n, dtype=dtype)
        comp3[1:] = 1 / (1 - z1)

        expected_fwd = comp1 + comp2 + comp3

        # Reverse condition
        sym_cond = -c0 / (z1 - 1.0) * expected_fwd[-1]

        # Backward pass
        # The transfer function for the forward result is equivalent to
        # the forward system times c0 / (1 - z1 * z).

        # Computing a closed form for the complete expression is difficult
        # The result will be computed iteratively from the difference equation
        exp_out = np.zeros(n, dtype=dtype)
        exp_out[0] = sym_cond

        for i in range(1, n):
            exp_out[i] = c0 * expected_fwd[n - 1 - i] + z1 * exp_out[i - 1]

        exp_out = exp_out[::-1]

        out = symiirorder1(signal, c0, z1, precision)
        xp_assert_close(out, exp_out, atol=4e-6, rtol=6e-7)

    @pytest.mark.parametrize('dtype', ['float32', 'float64'])
    def test_symiir1_values(self, dtype):
        rng = np.random.RandomState(1234)
        dtype = getattr(np, dtype)
        s = rng.uniform(size=16).astype(dtype)
        res = symiirorder1(s, 0.5, 0.1)

        # values from scipy 1.9.1
        exp_res = np.array([0.14387447, 0.35166047, 0.29735238, 0.46295986, 0.45174927,
                            0.19982875, 0.20355805, 0.47378628, 0.57232247, 0.51597393,
                           0.25935107, 0.31438554, 0.41096728, 0.4190693 , 0.25812255,
                           0.33671467], dtype=res.dtype)
        assert res.dtype == dtype
        atol = {np.float64: 1e-15, np.float32: 1e-7}[dtype]
        xp_assert_close(res, exp_res, atol=atol)

        s = s + 1j*s
        res = symiirorder1(s, 0.5, 0.1)
        assert res.dtype == np.complex64 if dtype == np.float32 else np.complex128
        xp_assert_close(res, exp_res + 1j*exp_res, atol=atol)

    @pytest.mark.parametrize(
        'dtype', ['float32', 'float64'])
    @pytest.mark.parametrize('precision', [-1.0, 0.7, 0.5, 0.25, 0.0075])
    def test_symiir2_initial_fwd(self, dtype, precision):
        dtype = getattr(np, dtype)
        c_precision = precision
        if precision <= 0.0 or precision > 1.0:
            if dtype in {np.float32, np.complex64}:
                c_precision = 1e-6
            else:
                c_precision = 1e-11

        # Compute the initial conditions for a order-two symmetrical low-pass
        # filter with r = 0.5 and omega = pi / 3 for an unit step input.
        r = np.asarray(0.5, dtype=dtype)
        omega = np.asarray(np.pi / 3.0, dtype=dtype)
        cs = 1 - 2 * r * np.cos(omega) + r**2

        # The index n for the initial condition is bound from 0 to the
        # first position where sin(omega * (n + 2)) = 0 => omega * (n + 2) = pi
        # For omega = pi / 3, the maximum initial condition occurs when
        # sqrt(3) / 2 * r**n < precision.
        # => n = log(2 * sqrt(3) / 3 * precision) / log(r)
        ub = np.ceil(np.log(c_precision / np.sin(omega)) / np.log(c_precision))
        lb = np.ceil(np.pi / omega) - 2
        n_exp = min(ub, lb)

        # The forward initial condition for a filter of order two is:
        # \frac{cs}{\sin(\omega)} \sum_{n = 0}^{N - 1} {
        #    r^(n + 1) \sin{\omega(n + 2)}} + cs
        # The closed expression for this sum is:
        # s[n] = 2 * r * np.cos(omega) -
        #        r**2 - r**(n + 2) * np.sin(omega * (n + 3)) / np.sin(omega) +
        #        r**(n + 3) * np.sin(omega * (n + 2)) / np.sin(omega) + cs
        fwd_initial_1 = (
            cs +
            2 * r * np.cos(omega) -
            r**2 -
            r**(n_exp + 2) * np.sin(omega * (n_exp + 3)) / np.sin(omega) +
            r**(n_exp + 3) * np.sin(omega * (n_exp + 2)) / np.sin(omega))

        # The second initial condition is given by
        # s[n] = 1 / np.sin(omega) * (
        #        r**2 * np.sin(3 * omega) -
        #        r**3 * np.sin(2 * omega) -
        #        r**(n + 3) * np.sin(omega * (n + 4)) +
        #        r**(n + 4) * np.sin(omega * (n + 3)))
        ub = np.ceil(np.log(c_precision / np.sin(omega)) / np.log(c_precision))
        lb = np.ceil(np.pi / omega) - 3
        n_exp = min(ub, lb)

        fwd_initial_2 = (
            cs + cs * 2 * r * np.cos(omega) +
            (r**2 * np.sin(3 * omega) -
             r**3 * np.sin(2 * omega) -
             r**(n_exp + 3) * np.sin(omega * (n_exp + 4)) +
             r**(n_exp + 4) * np.sin(omega * (n_exp + 3))) / np.sin(omega))

        expected = np.r_[fwd_initial_1, fwd_initial_2][None, :]
        expected = expected.astype(dtype)

        n = 100
        signal = np.ones(n, dtype=dtype)

        out = symiirorder2_ic_fwd(signal, r, omega, precision)
        xp_assert_close(out, expected, atol=4e-6, rtol=6e-7)

    @pytest.mark.parametrize(
        'dtype', [np.float32, np.float64])
    @pytest.mark.parametrize('precision', [-1.0, 0.7, 0.5, 0.25, 0.0075])
    def test_symiir2_initial_bwd(self, dtype, precision):
        c_precision = precision
        if precision <= 0.0 or precision > 1.0:
            if dtype in {np.float32, np.complex64}:
                c_precision = 1e-6
            else:
                c_precision = 1e-11

        r = np.asarray(0.5, dtype=dtype)
        omega = np.asarray(np.pi / 3.0, dtype=dtype)
        cs = 1 - 2 * r * np.cos(omega) + r * r
        a2 = 2 * r * np.cos(omega)
        a3 = -r * r

        n = 100
        signal = np.ones(n, dtype=dtype)

        # Compute initial forward conditions
        ic = symiirorder2_ic_fwd(signal, r, omega, precision)
        out = np.zeros(n + 2, dtype=dtype)
        out[:2] = ic[0]

        # Apply the forward system cs / (1 - a2 * z^-1 - a3 * z^-2))
        for i in range(2, n + 2):
            out[i] = cs * signal[i - 2] + a2 * out[i - 1] + a3 * out[i - 2]

        # Find the backward initial conditions
        ic2 = np.zeros(2, dtype=dtype)
        idx = np.arange(n)

        diff = (_compute_symiirorder2_bwd_hs(idx, cs, r * r, omega) +
                _compute_symiirorder2_bwd_hs(idx + 1, cs, r * r, omega))
        ic2_0_all = np.cumsum(diff * out[:1:-1])
        pos = np.where(diff ** 2 < c_precision)[0]
        ic2[0] = ic2_0_all[pos[0]]

        diff = (_compute_symiirorder2_bwd_hs(idx - 1, cs, r * r, omega) +
                _compute_symiirorder2_bwd_hs(idx + 2, cs, r * r, omega))
        ic2_1_all = np.cumsum(diff * out[:1:-1])
        pos = np.where(diff ** 2 < c_precision)[0]
        ic2[1] = ic2_1_all[pos[0]]

        out_ic = symiirorder2_ic_bwd(out, r, omega, precision)[0]
        xp_assert_close(out_ic, ic2, atol=4e-6, rtol=6e-7)

    @pytest.mark.parametrize(
        'dtype', [np.float32, np.float64])
    @pytest.mark.parametrize('precision', [-1.0, 0.7, 0.5, 0.25, 0.0075])
    def test_symiir2(self, dtype, precision):
        r = np.asarray(0.5, dtype=dtype)
        omega = np.asarray(np.pi / 3.0, dtype=dtype)
        cs = 1 - 2 * r * np.cos(omega) + r * r
        a2 = 2 * r * np.cos(omega)
        a3 = -r * r

        n = 100
        signal = np.ones(n, dtype=dtype)

        # Compute initial forward conditions
        ic = symiirorder2_ic_fwd(signal, r, omega, precision)
        out1 = np.zeros(n + 2, dtype=dtype)
        out1[:2] = ic[0]

        # Apply the forward system cs / (1 - a2 * z^-1 - a3 * z^-2))
        for i in range(2, n + 2):
            out1[i] = cs * signal[i - 2] + a2 * out1[i - 1] + a3 * out1[i - 2]

        # Find the backward initial conditions
        ic2 = symiirorder2_ic_bwd(out1, r, omega, precision)[0]

        # Apply the system cs / (1 - a2 * z - a3 * z^2)) in backwards
        exp = np.empty(n, dtype=dtype)
        exp[-2:] = ic2[::-1]

        for i in range(n - 3, -1, -1):
            exp[i] = cs * out1[i] + a2 * exp[i + 1] + a3 * exp[i + 2]

        out = symiirorder2(signal, r, omega, precision)
        xp_assert_close(out, exp, atol=4e-6, rtol=6e-7)

    @pytest.mark.parametrize('dtyp', ['float32', 'float64'])
    def test_symiir2_values(self, dtyp):
        dtyp = getattr(np, dtyp)
        rng = np.random.RandomState(1234)
        s = rng.uniform(size=16).astype(dtyp)
        res = symiirorder2(s, 0.1, 0.1, precision=1e-10)

        # values from scipy 1.9.1
        exp_res = np.array([0.26572609, 0.53408018, 0.51032696, 0.72115829, 0.69486885,
           0.3649055 , 0.37349478, 0.74165032, 0.89718521, 0.80582483,
           0.46758053, 0.51898709, 0.65025605, 0.65394321, 0.45273595,
           0.53539183], dtype=dtyp)

        assert res.dtype == dtyp
        # The values in SciPy 1.14 agree with those in SciPy 1.9.1 to this
        # accuracy only. Implementation differences are twofold:
        # 1. boundary conditions are computed differently
        # 2. the filter itself uses sosfilt instead of a hardcoded iteration
        # The boundary conditions seem are tested separately (see
        # test_symiir2_initial_{fwd,bwd} above, so the difference is likely
        # due to a different way roundoff errors accumulate in the filter.
        # In that respect, sosfilt is likely doing a better job.
        xp_assert_close(res, exp_res, atol=2e-6)

        s = s + 1j*s
        with pytest.raises(TypeError):
            res = symiirorder2(s, 0.5, 0.1)

    def test_symiir1_integer_input(self):
        s = np.where(np.arange(100) % 2, -1, 1)
        expected = symiirorder1(s.astype(float), 0.5, 0.5)
        out = symiirorder1(s, 0.5, 0.5)
        xp_assert_close(out, expected)

    def test_symiir2_integer_input(self):
        s = np.where(np.arange(100) % 2, -1, 1)
        expected = symiirorder2(s.astype(float), 0.5, np.pi / 3.0)
        out = symiirorder2(s, 0.5, np.pi / 3.0)
        xp_assert_close(out, expected)