FIR filter design using the window method.

Computes the coefficients of a finite impulse response (FIR) filter. The filter has linear phase (Type I for odd num_taps, Type II for even). Type II filters have a zero at Nyquist, so a filter requiring gain there must use an odd number of taps.

arguments

Arguments

• num_taps - number of filter coefficients (filter order + 1).
• cutoff - list of cutoff frequencies in the same units as sampling_rate.

options

Options

• :window - window function to apply. One of :hamming, :hann, :blackman, :bartlett, :rectangular, or {:kaiser, beta}. Defaults to :hamming.
• :pass_zero - if true, the DC gain is 1 (lowpass or bandstop); if false, the DC gain is 0 (highpass or bandpass). Defaults to true.
• :scale - if true, normalise the coefficients so the frequency response is exactly 1 at a reference frequency. Defaults to true.
• :sampling_rate - the sampling rate in Hz; cutoff is given in the same units. Defaults to 2.0 so cutoffs are already normalised to [0, 1] where 1 equals Nyquist.
• :type - output tensor type. Defaults to {:f, 32}.

examples

Examples

iex> coeffs = NxSignal.Filters.firwin(5, [0.3], window: :hamming, sampling_rate: 2.0)
iex> Nx.shape(coeffs)
{5}

Performs a median filter on a tensor.

options

Options

• :kernel_shape - the shape of the sliding window. It must be compatible with the shape of the tensor.

Applies a Wiener filter to the given Nx tensor.

options

Options

* `:kernel_size` - filter size given either a number or a tuple. 
  If a number is given, a kernel with the given size, and same number of axes 
  as the input tensor will be used. Defaults to `3`.
* `:noise` - noise power, given as a scalar. This will be estimated based on the input tensor if `nil`. Defaults to `nil`.

examples

Examples

iex> t = Nx.tensor([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]])
iex> NxSignal.Filters.wiener(t, kernel_size: {2, 2}, noise: 10)
#Nx.Tensor<
  f32[3][3]
  [
    [0.25, 0.75, 1.25],
    [1.25, 3.0, 4.0],
    [2.75, 6.0, 7.0]
  ]
>