| 8 | |
| 9 | |
| 10 | class FilteredSource(CustomSource): |
| 11 | def __init__( |
| 12 | self, |
| 13 | center_frequency: float, |
| 14 | frequencies: np.ndarray, |
| 15 | frequency_response, |
| 16 | dt, |
| 17 | time_src=None, |
| 18 | ): |
| 19 | # divide by two to compensate for staggered E, H time interval |
| 20 | dt = dt / 2 |
| 21 | self.dt = dt |
| 22 | self.frequencies = frequencies |
| 23 | self.center_frequencies = frequencies |
| 24 | |
| 25 | # For now, the basis functions cannot overlap much in the frequency |
| 26 | # domain. Otherwise, the resulting nodes are wildly large and induce |
| 27 | # numerical precision errors. We can always produce a safe result |
| 28 | # by forcing the length of each basis function to meet the minimum |
| 29 | # frequency requirements. This method still minimizes storage |
| 30 | # requirements. |
| 31 | self.T = np.max(np.abs(1 / np.diff(frequencies))) |
| 32 | self.N = np.rint(self.T / self.dt) |
| 33 | self.t = np.arange(0, dt * (self.N), dt) |
| 34 | self.n = np.arange(self.N) |
| 35 | f = self.func() |
| 36 | |
| 37 | # frequency bandwidth of the Nuttall window function |
| 38 | fwidth = self.nuttall_bandwidth() |
| 39 | |
| 40 | self.bf = [ |
| 41 | lambda t, i=i: 0 |
| 42 | if t > self.T |
| 43 | else ( |
| 44 | self.nuttall(t, self.center_frequencies) |
| 45 | / (self.dt / np.sqrt(2 * np.pi)) |
| 46 | )[i] |
| 47 | for i in range(len(self.center_frequencies)) |
| 48 | ] |
| 49 | self.time_src_bf = [ |
| 50 | CustomSource( |
| 51 | src_func=bfi, |
| 52 | center_frequency=center_frequency, |
| 53 | is_integrated=False, |
| 54 | end_time=self.T, |
| 55 | fwidth=fwidth, |
| 56 | ) |
| 57 | for bfi in self.bf |
| 58 | ] |
| 59 | |
| 60 | if time_src: |
| 61 | # get the cutoff of the input signal |
| 62 | signal_t = np.array( |
| 63 | [time_src.swigobj.current(ti, dt) for ti in self.t] |
| 64 | ) # time domain signal |
| 65 | signal_dtft = self.dtft(signal_t, self.frequencies) |
| 66 | else: |
| 67 | signal_dtft = 1 |
no outgoing calls
no test coverage detected