Skip to content

fir_filter

FIR Filter Layer API

This module provides classes to build FIR filter layers.

Classes:

Classes

FirFilter 

FirFilter(b: npt.NDArray[np.float32], a: npt.NDArray[np.float32] | None = None, forward_backward: bool = False, **kwargs)

Apply FIR filter to the input.

Parameters:

  • b (ndarray) –

    Numerator coefficients.

  • a (ndarray | None, default: None ) –

    Denominator coefficients. Defaults to None.

  • forward_backward (bool, default: False ) –

    Apply filter forward and backward. Defaults to False.

Example:

# Create sine wave at 10 Hz with 1000 Hz sampling rate
t = np.linspace(0, 1, 1000, endpoint=False)
x = np.sin(2 * np.pi * 10 * t)
# Add noise at 100 Hz and 2 Hz
x_noise = x + 0.5 * np.sin(2 * np.pi * 100 * t) + 0.5 * np.sin(2 * np.pi * 2 * t)
x_noise = x_noise.reshape(-1, 1).astype(np.float32)
x_noise = keras.ops.convert_to_tensor(x_noise)
# Create bandpass FIR taps
b = scipy.signal.firwin(101, [5, 15], fs=1000, pass_zero=False)
lyr = nse.layers.preprocessing.FirFilter(b, forward_backward=True)
y = lyr(x_noise).numpy().squeeze()
x_noise = x_noise.numpy().squeeze()
plt.plot(x, label="Original")
plt.plot(x_noise, label="Noisy")
plt.plot(y, label="Filtered")
plt.legend()
plt.show()
Source code in neuralspot_edge/layers/preprocessing/fir_filter.py 
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
def __init__(
    self,
    b: npt.NDArray[np.float32],
    a: npt.NDArray[np.float32] | None = None,
    forward_backward: bool = False,
    **kwargs,
):
    """Apply FIR filter to the input.

    Args:
        b (np.ndarray): Numerator coefficients.
        a (np.ndarray|None): Denominator coefficients. Defaults to None.
        forward_backward (bool): Apply filter forward and backward. Defaults to False.

    Example:

    ```python
    # Create sine wave at 10 Hz with 1000 Hz sampling rate
    t = np.linspace(0, 1, 1000, endpoint=False)
    x = np.sin(2 * np.pi * 10 * t)
    # Add noise at 100 Hz and 2 Hz
    x_noise = x + 0.5 * np.sin(2 * np.pi * 100 * t) + 0.5 * np.sin(2 * np.pi * 2 * t)
    x_noise = x_noise.reshape(-1, 1).astype(np.float32)
    x_noise = keras.ops.convert_to_tensor(x_noise)
    # Create bandpass FIR taps
    b = scipy.signal.firwin(101, [5, 15], fs=1000, pass_zero=False)
    lyr = nse.layers.preprocessing.FirFilter(b, forward_backward=True)
    y = lyr(x_noise).numpy().squeeze()
    x_noise = x_noise.numpy().squeeze()
    plt.plot(x, label="Original")
    plt.plot(x_noise, label="Noisy")
    plt.plot(y, label="Filtered")
    plt.legend()
    plt.show()
    ```
    """
    super().__init__(**kwargs)
    b = b.reshape(-1, 1, 1)
    self.b = self.add_weight(
        name="b",
        shape=b.shape,
        trainable=False,
    )
    self.b.assign(b)
    if a is not None:
        a = a.reshape(-1, 1, 1)
        self.a = self.add_weight(
            name="a",
            shape=a.shape,
            trainable=False,
        )
        self.a.assign(a)
    else:
        self.a = None

    self.forward_backward = forward_backward

Functions

augment_samples 
augment_samples(inputs) -> keras.KerasTensor

Augment a batch of samples during training.

Source code in neuralspot_edge/layers/preprocessing/fir_filter.py 
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
def augment_samples(self, inputs) -> keras.KerasTensor:
    """Augment a batch of samples during training."""

    samples = inputs[self.SAMPLES]

    # TODO: Handle 'a' coefficients
    if self.a is not None:
        raise NotImplementedError("Denominator coefficients 'a' are not supported yet.")

    outputs = keras.ops.depthwise_conv(samples, self.b, padding="same")

    if self.forward_backward:
        outputs = keras.ops.flip(outputs, axis=self.data_axis)
        outputs = keras.ops.depthwise_conv(outputs, self.b, padding="same")
        outputs = keras.ops.flip(outputs, axis=self.data_axis)
    # END IF
    return outputs
get_config 
get_config()

Serialize the layer configuration.

Source code in neuralspot_edge/layers/preprocessing/fir_filter.py 
 96
 97
 98
 99
100
101
102
103
104
def get_config(self):
    """Serialize the layer configuration."""
    config = super().get_config()
    config.update(
        {
            "forward_backward": self.forward_backward,
        }
    )
    return config