simulations/models/optical_channel.py

import matplotlib.pyplot as plt

import defs
import numpy as np
import math
from scipy.fft import fft, ifft


class OpticalChannel(defs.Channel):
    def __init__(self, noise_level, dispersion, symbol_rate, sample_rate, length, show_graphs=False, **kwargs):
        """
        :param noise_level: Noise level in dB
        :param dispersion: dispersion coefficient is ps^2/km
        :param symbol_rate: Symbol rate of modulated signal in Hz
        :param sample_rate: Sample rate of time-domain model (time steps in simulation) in Hz
        :param length: fibre length in km
        :param show_graphs: if graphs should be displayed or not

        Optical Channel class constructor
        """
        super().__init__(**kwargs)
        self.noise = 10 ** (noise_level / 10)

        self.dispersion = dispersion
        self.symbol_rate = symbol_rate
        self.symbol_period = 1 / self.symbol_rate
        self.sample_rate = sample_rate
        self.sample_period = 1 / self.sample_rate
        self.length = length
        self.show_graphs = show_graphs

    def __get_time_domain(self, symbol_vals):
        samples_per_symbol = int(self.sample_rate / self.symbol_rate)
        samples = int(symbol_vals.shape[0] * samples_per_symbol)

        symbol_vals_a = np.repeat(symbol_vals, repeats=samples_per_symbol, axis=0)
        t = np.linspace(start=0, stop=samples * self.sample_period, num=samples)
        val_t = symbol_vals_a[:, 0] * np.cos(2 * math.pi * symbol_vals_a[:, 2] * t + symbol_vals_a[:, 1])

        return t, val_t

    def __time_to_frequency(self, values):
        val_f = fft(values)
        f = np.linspace(0.0, 1 / (2 * self.sample_period), (values.size // 2))
        f_neg = -1 * np.flip(f)
        f = np.concatenate((f, f_neg), axis=0)
        return f, val_f

    def __frequency_to_time(self, values):
        val_t = ifft(values)
        t = np.linspace(start=0, stop=values.size * self.sample_period, num=values.size)
        return t, val_t

    def __apply_dispersion(self, values):
        # Obtain fft
        f, val_f = self.__time_to_frequency(values)

        if self.show_graphs:
            plt.plot(f, val_f)
            plt.title('frequency domain (pre-distortion)')
            plt.show()

        np.savetxt("foo.csv", f, delimiter=",")
        np.savetxt("barr.csv", np.real(val_f), delimiter=",")
        np.savetxt("bari.csv", np.imag(val_f), delimiter=",")

        # Apply distortion
        dist_val_f = val_f * np.exp(0.5j * self.dispersion * self.length * np.power(2 * math.pi * f, 2))

        if self.show_graphs:
            plt.plot(f, dist_val_f)
            plt.title('frequency domain (post-distortion)')
            plt.show()

        # Inverse fft
        t, val_t = self.__frequency_to_time(dist_val_f)

        return t, val_t

    def forward(self, values):
        # Converting APF representation to time-series
        t, val_t = self.__get_time_domain(values)

        if self.show_graphs:
            plt.plot(t, val_t)
            plt.title('time domain (raw)')
            plt.show()

        # Adding AWGN
        val_t += np.random.normal(0, 1, val_t.shape) * self.noise

        if self.show_graphs:
            plt.plot(t, val_t)
            plt.title('time domain (AWGN)')
            plt.show()

        # Applying chromatic dispersion
        t, val_t = self.__apply_dispersion(val_t)

        if self.show_graphs:
            plt.plot(t, val_t)
            plt.title('time domain (post-distortion)')
            plt.show()

        return t, val_t


if __name__ == '__main__':
    # Simple OOK modulation
    num_of_symbols = 10
    symbol_vals = np.zeros((num_of_symbols, 3))

    symbol_vals[:, 0] = np.random.randint(2, size=symbol_vals.shape[0])
    symbol_vals[:, 2] = 10e6

    channel = OpticalChannel(noise_level=-20, dispersion=-21.7, symbol_rate=100e3,
                             sample_rate=500e6, length=100, show_graphs=True)
    time, v = channel.forward(symbol_vals)