cbadc.digital_estimator.adaptive_filter.AdaptiveIIRFilter

class cbadc.digital_estimator.adaptive_filter.AdaptiveIIRFilter(M, K, L=1, dtype=<class 'numpy.float64'>)[source]

Bases: AdaptiveFIRFilter

The adaptive IIR filter model.

Methods

`__init__`(M, K[, L, dtype])
`call`(x)	param x The input data, shape (nr_samples, M - nr_references).
`get_filter`()	Returns the FIR filter.
`get_offset`()	Returns the offset.
`gradient`(x, y)	Computes the gradient of the loss function with respect to the filter coefficients.
`impulse_response`([number_of_points])	Returns the impulse response of the filter.
`lms`(x, y[, delay])	Fits the filter to the given data using the LMS method.
`loss`(x, y)	Computes the loss function for the given FIR filter.
`lstsq`(x, y[, delay, x_validate, verbose])	Fits the filter to the given data using the least squares method.
`plot_bode`([linear_frequency])	Plots the Bode diagram for the filter.
`plot_impulse_response`()	Plots the impulse response of the filter for each channel.
`predict`(x)	param x The input data batch.
`predict_full`(x, y)	In contrast to the predict method, this method additionally removes the reference signals from the output.
`rls`(x, y[, delay])	Fits the filter to the given data using the RLS method.
`transfer_function`([number_of_points])	Returns the transfer function of the filter.

call(x: ndarray)

Parameters: x (np.ndarray (batch_size, M, K)) – The input data, shape (nr_samples, M - nr_references).
Returns: y – The output data, shape (nr_samples, nr_references).
Return type: np.ndarray (L, batch_size)

get_filter()

Returns the FIR filter.

Returns: h – The FIR filter.
Return type: np.ndarray (L, M, K)

get_offset()

Returns the offset.

Returns: offset – The offset.
Return type: np.ndarray (L,)

gradient(x: ndarray, y: ndarray) → List[ndarray]

Computes the gradient of the loss function with respect to the filter coefficients.

Parameters

x (np.ndarray (batch_size, M, K)) – The input data.
y (np.ndarray (L, batch_size)) – The output data.

Returns

gradient – The gradient of the loss function with respect to the filter coefficients.

Return type

[np.ndarray (L, M, K), np.ndarray (L,)]

impulse_response(number_of_points: int = 0)

Returns the impulse response of the filter.

Parameters: number_of_points (int) – The number of uniformly spaced frequency points to evaluate the transfer function at, defaults to K.
Returns: h – The impulse response.
Return type: np.ndarray (L, M, K)

lms(x: ndarray, y: ndarray, delay: int = 0, **kwargs)[source]

Fits the filter to the given data using the LMS method.

Parameters

x (np.ndarray (batch_size, M)) – The input data.
y (np.ndarray (L, batch_size)) – The reference data.
delay (int) – The delay of the IIR filter. must be non-negative.
batch_size (int) – The batch size.
epochs (int) – The number of epochs.
learning_rate (float) – The learning rate, defaults to 1e-5.
momentum (float) – The momentum, defaults to 0.9.
verbose (bool) – Whether to print the loss function during training.

Returns

loss – The loss function evaluated on the given data.

Return type

np.ndarray (L,)

loss(x: ndarray, y: ndarray)

Computes the loss function for the given FIR filter.

Parameters

x (np.ndarray (batch_size, M, K)) – The input data.
y (np.ndarray (L, batch_size)) – The reference data.

Returns

loss – The loss function evaluated on the given data.

Return type

np.ndarray (L,)

lstsq(x: ndarray, y: ndarray, delay: int = 0, x_validate: ndarray = None, verbose=True)[source]

Fits the filter to the given data using the least squares method.

Parameters

x (np.ndarray (batch_size, M)) – The input data.
y (np.ndarray (L, batch_size)) – The reference data.
delay (int) – The delay of the IIR filter. must be non-negative.
verbose (bool) – Whether to print the loss function during training.

Returns

loss – The loss function evaluated on the given data.

Return type

np.ndarray (L,)

plot_bode(linear_frequency: bool = False)

Plots the Bode diagram for the filter.

Mint: The number of channels.
linear_frequencybool: If true, the frequency axis is linear, otherwise it is logarithmic.

f_hmatplotlib.figure.Figure: The figure object.
ax_hnumpy.ndarray: The axes objects.

plot_impulse_response()

Plots the impulse response of the filter for each channel.

Mint: The number of channels.

f_hmatplotlib.figure.Figure: The figure object.
ax_hnumpy.ndarray: The axes objects.

predict(x: ndarray)[source]

Parameters: x (np.ndarray (batch_size, M)) – The input data batch.
Returns: y – The output data, shape (nr_samples, nr_references).
Return type: np.ndarray (L, batch_size)

predict_full(x, y)

In contrast to the predict method, this method additionally removes the reference signals from the output.

Parameters

x (np.ndarray [batch_size, M, K]) – The input data, shape (nr_samples, M - nr_references).
y (np.ndarray [L, batch_size]) – The reference data.

rls(x: ndarray, y: ndarray, delay: int = 0, **kwargs)[source]

Fits the filter to the given data using the RLS method.

Parameters

x (np.ndarray (batch_size, M)) – The input data.
y (np.ndarray (L, batch_size)) – The reference data.
delay (int) – The delay of the IIR filter. must be non-negative.
epochs (int) – The number of epochs.
delta (float) – The delta parameter of the RLS algorithm.
lambda (float) – The lambda parameter of the RLS algorithm.
verbose (bool) – Whether to print the loss function during training.

Returns

loss – The loss function evaluated on the given data.

Return type

np.ndarray (L,)

transfer_function(number_of_points: int = 0)[source]

Returns the transfer function of the filter.

Parameters

number_of_points (int) – The number of uniformly spaced frequency points to evaluate the transfer function at, defaults to K.

Returns

freqs (np.ndarray) – The frequency points.
tf (np.ndarray (L, M, number_of_points))