mokka.pulseshaping

Module implementing pulseshaping.

Module implementing pulseshaping with PyTorch.

class mokka.pulseshaping.torch.MidriseQuantizer(bit, max_amplitude=1.0)

Bases: Module

A Quantizer which translates a continuos signal into 2**m levels.

This quantizer operates on a real-valued signal and quantizes between [-max_amplitude,max_amplitude].

__init__(bit, max_amplitude=1.0)

Construct MidriseQuuantizer.

Parameters:

  • bit – number of input bits, this will give 2**bit quantization levels

  • max_amplitude – Maximum amplitude before saturation.

forward(y)

Perform quantization.

Parameters:

y – real-valued signal to quantize

class mokka.pulseshaping.torch.PulseShaping(impulse_response: tensor, learnable: bool = False)

Bases: Module

PulseShaping filter class.

Several classical shapes, a filter function and a matched filter function are provided.

__init__(impulse_response: tensor, learnable: bool = False) → None

Method generated by attrs for class PulseShaping.

forward(y, n_up)

Perform pulse shaping.

Parameters:

  • y – transmit signal

  • n_up – upsampling factor

Returns:

shaped and upsampled transmit signal at sample rate

classmethod get_rc_ir(syms, r, n_up, learnable=False, normalize=True, normalize_symbol=False)

Determine normed coefficients of an RC filter.

Formula out of: K.-D. Kammeyer, Nachrichtenübertragung At poles, l’Hospital was used

Parameters:

  • syms – “normed” length of ir. ir-length will be 2*syms+1

  • r – roll-off factor [Float]

  • n_up – upsampling factor [Int]

Returns:

tuple containing time-index-array and impulse response in an array

classmethod get_rrc_ir(syms, r, n_up, learnable=False, normalize=True, normalize_symbol=False)

Determine normed coefficients of an RRC filter.

This function is adapted from Sourcecode written by Dominik Rimpf and was published under the MIT license [https://gitlab.com/domrim/bachelorarbeit-code]

Formula out of: K.-D. Kammeyer, Nachrichtenübertragung At poles, l’Hospital was used

Parameters:

  • syms – “normed” length of ir. ir-length will be 2*syms+1

  • r – roll-off factor [Float]

  • n_up – upsampling factor [Int]

Returns:

tuple containing time-index-array and impulse response in an array

impulse_response_conj: tensor
learnable: bool
matched(r, n_down, max_energy=False)

Perform matched filtering.

This function assumes perfect timing sync.

Parameters:

  • r – received complex signal

  • n_down – downsampling factor

Returns:

filtered and downsampled signal at symbol rate

normalize_filter()

Normalize filter taps saved in this object.

class mokka.pulseshaping.torch.SoftMidriseQuantizer(bit, max_amplitude=1.0, temperature=0.01)

Bases: Module

Use tanh() with temperature to approximate Midrise quantization.

For temperature -> 0 the SoftMidriseQuantizer approximates the MidriseQuantizer.

__init__(bit, max_amplitude=1.0, temperature=0.01)

Initialize SoftMidriseQuantizer.

Parameters:

  • bit – number of bits to quantize. This will give 2**bit quantization levels.

  • max_amplitude – Maximum amplitude before saturation.

  • temperature – scale input to tanh to achieve a steeper quantization.

forward(x)

Apply quantization to signal.

Parameters:

x – real signal to quantize.

Returns:

Soft quantized signal.

mokka.pulseshaping.torch.blackman_harris_window(length=501)

Get Blackman-harris window coefficients.

Parameters:

length – number of window taps

mokka.pulseshaping.torch.brickwall_filter(cutoff, filter_length, filter_gain, window='hann', dtype=torch.complex64)

Approximate brickwall filter in fourier domain with configurable window.

mokka.pulseshaping.torch.cosine_sum_window(coeffs, length=501)

Get cosine sum window coefficients.

Parameters:

coeffs – \(a_k\) coefficients to calculate cosine sum window

mokka.pulseshaping.torch.downsample(n_down, signal, filter_length=501, filter_gain=1, window='blackmann_harris')

Perform downsampling on signal.

Parameters:

  • n_down – downsampling factor

  • signal – signal to downsample

  • filter_length – number of filter taps to use for the brickwall filter

  • filter_gain – gain to apply to brickwall filter

mokka.pulseshaping.torch.hamming_window(length=501)

Get Hamming window coefficients.

Parameters:

length – number of window taps

mokka.pulseshaping.torch.hann_window(length=501)

Get Hann window coefficients.

Parameters:

length – number of window taps

mokka.pulseshaping.torch.upsample(n_up, signal, filter_length=501, filter_gain=1)

Perform upsampling on signal.

Parameters:

  • n_up – upsampling factor

  • signal – signal to upsample

Filter_length:

number of filter taps of the brickwall interpolation filter

Filter_gain:

gain to apply to brickwall filter