Files
oopt-gnpy/examples/edfa_model/amplifier.py
2018-02-13 12:11:10 -05:00

302 lines
11 KiB
Python

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Mon Nov 27 12:32:04 2017
@author: briantaylor
"""
import numpy as np
from numpy import polyfit, polyval, mean
from utilities import lin2db, db2lin, itufs, freq2wavelength
import matplotlib.pyplot as plt
from scipy.constants import h
def noise_profile(nf, gain, ffs, df):
""" noise_profile(nf, gain, ffs, df) computes amplifier ase
:param nf: Noise figure in dB
:param gain: Actual gain calculated for the EDFA in dB units
:param ffs: A numpy array of frequencies
:param df: the reference bw in THz
:type nf: numpy.ndarray
:type gain: numpy.ndarray
:type ffs: numpy.ndarray
:type df: float
:return: the asepower in dBm
:rtype: numpy.ndarray
ASE POWER USING PER CHANNEL GAIN PROFILE
INPUTS:
NF_dB - Noise figure in dB, vector of length number of channels or
spectral slices
G_dB - Actual gain calculated for the EDFA, vector of length number of
channels or spectral slices
ffs - Center frequency grid of the channels or spectral slices in THz,
vector of length number of channels or spectral slices
dF - width of each channel or spectral slice in THz,
vector of length number of channels or spectral slices
OUTPUT:
ase_dBm - ase in dBm per channel or spectral slice
NOTE: the output is the total ASE in the channel or spectral slice. For
50GHz channels the ASE BW is effectively 0.4nm. To get to noise power in
0.1nm, subtract 6dB.
ONSR is usually quoted as channel power divided by
the ASE power in 0.1nm RBW, regardless of the width of the actual
channel. This is a historical convention from the days when optical
signals were much smaller (155Mbps, 2.5Gbps, ... 10Gbps) than the
resolution of the OSAs that were used to measure spectral power which
were set to 0.1nm resolution for convenience. Moving forward into
flexible grid and high baud rate signals, it may be convenient to begin
quoting power spectral density in the same BW for both signal and ASE,
e.g. 12.5GHz."""
h_mWThz = 1e-3 * h * (1e14)**2
nf_lin = db2lin(nf)
g_lin = db2lin(gain)
ase = h_mWThz * df * ffs * (nf_lin * g_lin - 1)
asedb = lin2db(ase)
return asedb
def gain_profile(dfg, dgt, Pin, gp, gtp):
"""
:param dfg: design flat gain
:param dgt: design gain tilt
:param Pin: channing input power profile
:param gp: Average gain setpoint in dB units
:param gtp: gain tilt setting
:type dfg: numpy.ndarray
:type dgt: numpy.ndarray
:type Pin: numpy.ndarray
:type gp: float
:type gtp: float
:return: gain profile in dBm
:rtype: numpy.ndarray
AMPLIFICATION USING INPUT PROFILE
INPUTS:
DFG - vector of length number of channels or spectral slices
DGT - vector of length number of channels or spectral slices
Pin - input powers vector of length number of channels or
spectral slices
Gp - provisioned gain length 1
GTp - provisioned tilt length 1
OUTPUT:
amp gain per channel or spectral slice
NOTE: there is no checking done for violations of the total output power
capability of the amp.
Ported from Matlab version written by David Boerges at Ciena.
Based on:
R. di Muro, "The Er3+ fiber gain coefficient derived from a dynamic
gain
tilt technique", Journal of Lightwave Technology, Vol. 18, Iss. 3,
Pp. 343-347, 2000.
"""
err_tolerance = 1.0e-11
simple_opt = True
# TODO make all values linear unit and convert to dB units as needed within
# this function.
nchan = list(range(len(Pin)))
# TODO find a way to use these or lose them. Primarily we should have a
# way to determine if exceeding the gain or output power of the amp
tot_in_power_db = lin2db(np.sum(db2lin(Pin)))
avg_gain_db = lin2db(mean(db2lin(dfg)))
# Linear fit to get the
p = polyfit(nchan, dgt, 1)
dgt_slope = p[0]
# Calculate the target slope- Currently assumes equal spaced channels
# TODO make it so that supports arbitrary channel spacing.
targ_slope = gtp / (len(nchan) - 1)
# 1st estimate of DGT scaling
dgts1 = targ_slope / dgt_slope
# when simple_opt is true code makes 2 attempts to compute gain and
# the internal voa value. This is currently here to provide direct
# comparison with original Matlab code. Will be removed.
# TODO replace with loop
if simple_opt:
# 1st estimate of Er gain & voa loss
g1st = dfg + dgt * dgts1
voa = lin2db(mean(db2lin(g1st))) - gp
# 2nd estimate of Amp ch gain using the channel input profile
g2nd = g1st - voa
pout_db = lin2db(np.sum(db2lin(Pin + g2nd)))
dgts2 = gp - (pout_db - tot_in_power_db)
# Center estimate of amp ch gain
xcent = dgts2
gcent = g1st - voa + dgt * xcent
pout_db = lin2db(np.sum(db2lin(Pin + gcent)))
gavg_cent = pout_db - tot_in_power_db
# Lower estimate of Amp ch gain
deltax = np.max(g1st) - np.min(g1st)
xlow = dgts2 - deltax
glow = g1st - voa + xlow * dgt
pout_db = lin2db(np.sum(db2lin(Pin + glow)))
gavg_low = pout_db - tot_in_power_db
# Upper gain estimate
xhigh = dgts2 + deltax
ghigh = g1st - voa + xhigh * dgt
pout_db = lin2db(np.sum(db2lin(Pin + ghigh)))
gavg_high = pout_db - tot_in_power_db
# compute slope
slope1 = (gavg_low - gavg_cent) / (xlow - xcent)
slope2 = (gavg_cent - gavg_high) / (xcent - xhigh)
if np.abs(gp - gavg_cent) <= err_tolerance:
dgts3 = xcent
elif gp < gavg_cent:
dgts3 = xcent - (gavg_cent - gp) / slope1
else:
dgts3 = xcent + (-gavg_cent + gp) / slope2
gprofile = g1st - voa + dgt * dgts3
else:
gprofile = None
return gprofile
if __name__ == '__main__':
plt.close('all')
fc = itufs(0.05)
lc = freq2wavelength(fc) / 1000
nchan = list(range(len(lc)))
df = np.array([0.05] * (nchan[-1] + 1))
# TODO remove path dependence
path = ''
"""
DFG_96: Design flat gain at each wavelength in the 96 channel 50GHz ITU
grid in dB. This can be experimentally determined by measuring the gain
at each wavelength using a full, flat channel (or ASE) load at the input.
The amplifier should be set to its maximum flat gain (tilt = 0dB). This
measurement captures the ripple of the amplifier. If the amplifier was
designed to be mimimum ripple at some other tilt value, then the ripple
reflected in this measurement will not be that minimum. However, when
the DGT gets applied through the provisioning of tilt, the model should
accurately reproduce the expected ripple at that tilt value. One could
also do the measurement at some expected tilt value and back-calculate
this vector using the DGT method. Alternatively, one could re-write the
algorithm to accept a nominal tilt and a tiled version of this vector.
"""
dfg_96 = np.loadtxt(path + 'DFG_96.txt')
"""maximum gain for flat operation - the amp in the data file was designed
for 25dB gain and has an internal VOA for setting the external gain
"""
avg_dfg = dfg_96.mean()
"""
DGT_96: This is the so-called Dynamic Gain Tilt of the EDFA in dB/dB. It
is the change in gain at each wavelength corresponding to a 1dB change at
the longest wavelength supported. The value can be obtained
experimentally or through analysis of the cross sections or Giles
parameters of the Er fibre. This is experimentally measured by changing
the gain of the amplifier above the maximum flat gain while not changing
the internal VOA (i.e. the mid-stage VOA is set to minimum and does not
change during the measurement). Note that the measurement can change the
gain by an arbitrary amount and divide by the gain change (in dB) which
is measured at the reference wavelength (the red end of the band).
"""
dgt_96 = np.loadtxt(path + 'DGT_96.txt')
"""
pNFfit3: Cubic polynomial fit coefficients to noise figure in dB
averaged across wavelength as a function of gain change from design flat:
NFavg = pNFfit3(1)*dG^3 + pNFfit3(2)*dG^2 pNFfit3(3)*dG + pNFfit3(4)
where
dG = GainTarget - average(DFG_96)
note that dG will normally be a negative value.
"""
nf_fitco = np.loadtxt(path + 'pNFfit3.txt')
"""NFR_96: Noise figure ripple in dB away from the average noise figure
across the band. This captures the wavelength dependence of the NF. To
calculate the NF across channels, one uses the cubic fit coefficients
with the external gain target to get the average nosie figure, NFavg and
then adds this to NFR_96:
NF_96 = NFR_96 + NFavg
"""
nf_ripple = np.loadtxt(path + 'NFR_96.txt')
# This is an example to set the provisionable gain and gain-tilt values
# Tilt is in units of dB/THz
gain_target = 20.0
tilt_target = -0.7
# calculate the NF for the EDFA at this gain setting
dg = gain_target - avg_dfg
nf_avg = polyval(nf_fitco, dg)
nf_96 = nf_ripple + nf_avg
# get the input power profiles to show
pch2d = np.loadtxt(path + 'Pchan2D.txt')
# Load legend and assemble legend text
pch2d_legend_data = np.loadtxt(path + 'Pchan2DLegend.txt')
pch2d_legend = []
for ea in pch2d_legend_data:
s = ''.join([chr(xx) for xx in ea.astype(dtype=int)]).strip()
pch2d_legend.append(s)
# assemble plot
axis_font = {'fontname': 'Arial', 'size': '16', 'fontweight': 'bold'}
title_font = {'fontname': 'Arial', 'size': '17', 'fontweight': 'bold'}
tic_font = {'fontname': 'Arial', 'size': '12'}
plt.rcParams["font.family"] = "Arial"
plt.figure()
plt.plot(nchan, pch2d.T, '.-', lw=2)
plt.xlabel('Channel Number', **axis_font)
plt.ylabel('Channel Power [dBm]', **axis_font)
plt.title('Input Power Profiles for Different Channel Loading',
**title_font)
plt.legend(pch2d_legend, loc=5)
plt.grid()
plt.ylim((-100, -10))
plt.xlim((0, 110))
plt.xticks(np.arange(0, 100, 10), **tic_font)
plt.yticks(np.arange(-110, -10, 10), **tic_font)
plt.figure()
ea = pch2d[1, :]
for ea in pch2d:
chgain = gain_profile(dfg_96, dgt_96, ea, gain_target, tilt_target)
pase = noise_profile(nf_96, chgain, fc, df)
pout = lin2db(db2lin(ea + chgain) + db2lin(pase))
plt.plot(nchan, pout, '.-', lw=2)
plt.title('Output Power with ASE for Different Channel Loading',
**title_font)
plt.xlabel('Channel Number', **axis_font)
plt.ylabel('Channel Power [dBm]', **axis_font)
plt.grid()
plt.ylim((-50, 10))
plt.xlim((0, 100))
plt.xticks(np.arange(0, 100, 10), **tic_font)
plt.yticks(np.arange(-50, 10, 10), **tic_font)
plt.legend(pch2d_legend, loc=5)
plt.show()