Moving estimate_filter_length into datasets module (#90)

torchsig/datasets/__init__.py

@@ -0,0 +1,19 @@
+import numpy as np
+
+
+def estimate_filter_length(
+    transition_bandwidth: float, attenuation_db: int = 72, sample_rate: float = 1.0
+) -> int:
+    # estimate the length of an FIR filter using harris' approximaion,
+    # N ~= (sampling rate/transition bandwidth)*(sidelobe attenuation in dB / 22)
+    # fred harris, Multirate Signal Processing for Communication Systems,
+    # Second Edition, p.59
+    filter_length = int(
+        np.round((sample_rate / transition_bandwidth) * (attenuation_db / 22))
+    )
+
+    # odd-length filters are desirable because they do not introduce a half-sample delay
+    if np.mod(filter_length, 2) == 0:
+        filter_length += 1
+
+    return filter_length

torchsig/datasets/synthetic.py

@@ -10,6 +10,7 @@
 from torchsig.utils.dataset import SignalDataset
 from torchsig.utils.types import SignalData, SignalDescription
 from torchsig.transforms.functional import IntParameter, FloatParameter
+from torchsig.datasets import estimate_filter_length
 
 
 def torchsig_convolve(
@@ -169,21 +170,6 @@ class DigitalModulationDataset(ConcatDataset):
 
     """
 
-    def estimate_filter_length(attenuation_db, sample_rate, transition_bandwidth):
-        # estimate the length of an FIR filter using harris' approximaion,
-        # N ~= (sampling rate/transition bandwidth)*(sidelobe attenuation in dB / 22)
-        # fred harris, Multirate Signal Processing for Communication Systems,
-        # Second Edition, p.59
-        filter_length = int(
-            np.round((sample_rate / transition_bandwidth) * (attenuation_db / 22))
-        )
-
-        # odd-length filters are desirable because they do not introduce a half-sample delay
-        if np.mod(filter_length, 2) == 0:
-            filter_length += 1
-
-        return filter_length
-
     def __init__(
         self,
         modulations: Optional[Union[List, Tuple]] = ("bpsk", "2gfsk"),
@@ -380,11 +366,8 @@ def _generate_samples(self, item: Tuple) -> np.ndarray:
             (self.iq_samples_per_symbol * len(symbols),), dtype=np.complex64
         )
         zero_padded[:: self.iq_samples_per_symbol] = symbols
-
-        # estimate total filter length for pulse shape
-        attenuation_db = 72  # sidelobe attenuation level, 72 dB -> 12 bit dynamic range
-        pulse_shape_filter_length = DigitalModulationDataset.estimate_filter_length(
-            attenuation_db, 1, signal_description.excess_bandwidth
+        pulse_shape_filter_length = estimate_filter_length(
+            signal_description.excess_bandwidth
         )
         pulse_shape_filter_span = int(
             (pulse_shape_filter_length - 1) / 2
@@ -993,18 +976,12 @@ def _generate_samples(self, item: Tuple) -> np.ndarray:
             # accept the cutoff-frequency of the filter as external
             # parameter, randomized as part of outer framework
             cutoff_frequency = bandwidth
-            # define the sidelobe levels of the filter (in dB)
-            attenuation_db = 72
-            # using a normalized sampling rate of 1 such that fs/2 = 1/2
-            sample_rate = 1
             # calculate transition bandwidth. a larger cutoff frequency requires
             # a smaller transition bandwidth, and a smaller cutoff frequency
             # allows for a larger transition bandwidth
-            transition_bandwidth = (sample_rate / 2 - (cutoff_frequency)) / 4
+            transition_bandwidth = (1.0 / 2 - (cutoff_frequency)) / 4
             # estimate number of taps needed to implement filter
-            num_taps = DigitalModulationDataset.estimate_filter_length(
-                attenuation_db, sample_rate, transition_bandwidth
-            )
+            num_taps = estimate_filter_length(transition_bandwidth)
 
             # design the filter
             taps = sp.firwin(
@@ -1013,7 +990,7 @@ def _generate_samples(self, item: Tuple) -> np.ndarray:
                 width=transition_bandwidth,
                 window=sp.get_window("blackman", num_taps),
                 scale=True,
-                fs=sample_rate,
+                fs=1,
             )
             # apply the filter
             modulated = torchsig_convolve(modulated, taps, gpu=self.use_gpu)

torchsig/datasets/wideband.py

@@ -39,6 +39,7 @@
     uniform_continuous_distribution,
     uniform_discrete_distribution,
 )
+from torchsig.datasets import estimate_filter_length
 
 
 class SignalBurst(SignalDescription):
@@ -89,9 +90,7 @@ def generate_iq(self):
         center = lower + bandwidth / 2
 
         # Filter noise
-        num_taps = int(
-            2 * np.ceil(50 * 2 * np.pi / bandwidth / 0.125 / 22)
-        )  # fred harris rule of thumb *2
+        num_taps = estimate_filter_length((0.5 - 0.02 * bandwidth) / 4)
         sinusoid = np.exp(2j * np.pi * center * np.linspace(0, num_taps - 1, num_taps))
         taps = signal.firwin(
             num_taps,
@@ -366,10 +365,7 @@ def generate_iq(self):
                 -int(self.num_iq_samples * self.duration * oversample) :
             ]
 
-            # Filter around center
-            num_taps = int(
-                2 * np.ceil(50 * 2 * np.pi / 0.5 / 0.125 / 22)
-            )  # fred harris rule of thumb * 2
+            num_taps = estimate_filter_length((0.5 - 0.02 / oversample) / 4)
 
             taps = signal.firwin(
                 num_taps,
@@ -472,9 +468,8 @@ def generate_iq(self):
         )
 
         # Filter around center
-        num_taps = int(
-            2 * np.ceil(50 * 2 * np.pi / 0.5 / 0.125 / 22)
-        )  # fred harris rule of thumb * 2
+        num_taps = estimate_filter_length((0.5 - 0.02 * 0.5) / 4)
+
         taps = signal.firwin(
             num_taps,
             0.5,
@@ -580,9 +575,8 @@ def generate_iq(self):
         )
 
         # Filter around center
-        num_taps = int(
-            2 * np.ceil(50 * 2 * np.pi / 0.5 / 0.125 / 22)
-        )  # fred harris rule of thumb * 2
+        num_taps = estimate_filter_length((0.5 - 0.5 * 0.02) / 4)
+
         taps = signal.firwin(
             num_taps,
             0.5,