cumtrapz has been replaced with cumulative_trapezoid in the scipy version currently required by GNPy.
Change-Id: I6790f7aa8d8e8d171faa48db4b20e6a93141c471
An approximated version of the GGN is implemented to reduce the computational time enabling fast multi-band transmission simulations
Change-Id: I2951a878aa04b5eb4a33ba86d626a788c4cbb100
The chromatic dispersion and dispersion slope can be provided as a single values evaluated at the fiber reference frequency or in a dictionary containing the dispersion values evaluated at multiple frequencies:
"dispersion": {"value": [], "frequency": []}
Change-Id: I81429484dd373cc49bd9baf013247782ba1912fd
The nonlinear coefficient can be expressed at the reference frequency and will be scaled in frequency using the scaling rule of the effective area
Change-Id: Id103b227472702776bda17ab0a2a120ecfbf7473
The evaluation of the eta matrix is reintroduced for nli evaluation and validation purposes. Also, the parameters for cut and pump are separated explicitelly.
Change-Id: Id3844fa8ba41a5d4f5a72d281d758136ee983f45
1. Effective area scaling along frequency is implemented by means of a technological model.
2. Raman gain coefficient is extended coherently, including the scaling due to the pump frequency.
Change-Id: I4e8b79697500ef0f73ba2f969713d9bdb3e9949c
Co-authored-by: Giacomo Borraccini <giacomo.borraccini@polito.it>
In the previous version, the position of the lumped losses were not
included in the result Raman profile. As the latter is then used to
evaluate the NLI, including the lumped loss positions is required for
accurate estimations.
Change-Id: I683f48ceb7139d1a8be03d2e7ca7e3abffecbe85
The code look as if it was trying to prevent direct instantiation of the
SimParams class. However, instance *creation* in Python is actually
handled via `__new__` which was not overridden. In addition, the
`get()` accessor was invoking `SimParams.__new__()` directly, which
meant that this class was instantiated each time it was needed.
Let's cut the boilerplate by getting rid of the extra step and just use
the regular constructor.
This patch doesn't change anything in actual observable behavior. I
still do not like this implicit singleton design pattern, but nuking
that will have to wait until some other time.
Change-Id: I3ca81bcd0042e91b4f6b7581879922611f18febe
In the previous version, when the values of the counter-propagating Raman pump profiles were flipped, the pumps resulted flipped also in frequency.
Change-Id: I66f7c2aff35c72f5dcb4fb11f7a82fe1df2ee3f2
Co-authored-by: Andrea D'Amico <andrea.damico@polito.it>
numpy.empty should not be used for initializing arrays without manually
setting values since it does not initialize entries. Use numpy.zeros
instead.
Signed-off-by: Jonas Mårtensson <jonas.martensson@ri.se>
Change-Id: I4e85eb39bdce00663c0cab9582ea7ae25eb90986
The counterpropagating power profile was initialized based on number of
copropagating frequencies, which caused simulation to crash when no
counterpropagating pumps were present.
Signed-off-by: Jonas Mårtensson <jonas.martensson@ri.se>
Change-Id: I685e5438fda06058f0757ff51fdd67bc68aa1352
The lumped losses are used in the computation of the loss/gain profile
through the fiber whether the Raman effect is considered or not. The
computed power profile is used to calculate the related NLI impairment.
Using the 'gn_model_analytic' method, the lumped losses are taken into
account as the contribution of an additional total loss at the end of
the fiber span. In case the 'ggn_spectrally_separated' is selected, the
method uses the computed power profile according to the specified z and
frequency arrays. The lumped losses are so considered within the NLI
power evolution along the fiber.
Change-Id: I73a6baa321aca4d041cafa180f47afed824ce267
Signed-off-by: Jan Kundrát <jan.kundrat@telecominfraproject.com>
In this change, the RamanSolver is completely restructured in order to obtain a simplified and faster solution of the Raman equation. Additionally, the inter-channel Raman effect can be evaluated also in the standard fiber, when no Raman pumping is present. The same is true for the GGN model.
The Raman pump parameter pumps_loss_coef has been removed as it was not used. The loss coefficient value evaluated at the pump frequency can be included within the fiber loss_coef parameter.
This change induces variations in some expected test results as the Raman profile solution is calculated by a completely distinct algorithm. Nevertheless, these variations are negligible being lower than 0.1dB.
Change-Id: Iaa40fbb23c555571497e1ff3bf19dbcbfcadf96b
Modification of the Fiber and the NliSolver in order to properly propagate the new definition of the spectral information taking advantage of the numpy array structures.
In the previous version, the propagation of the spectral information was implemented by means of for cycles over each channel, in turn.
In this change the propagation is applied directly on the newly defined spectral information attributes as numpy arrays.
Additional changes:
- Simplification of the FiberParameters and the NliParameters;
- Previous issues regarding the loss_coef definition along the frequency are solved;
- New test in test_science_utils.py verifing that the fiber propagation provides the correct values in case of a few cases of flex grid spectra.
Change-Id: Id71f36effba35fc3ed4bbf2481a3cf6566ccb51c
This change siplifies the structure of the simulation parameters,
removing the gnpy.science_utils.simulation layer, provides some
documentation of the parameters and define a mock fixture for testing in
safe mode.
Jan: while I'm not thrilled by this concept of hidden global state, we
agreed to let it in as a temporary measure (so as not to hold merging of
Andrea's flexgrid/multirate patches). I've refactored this to a more
pytest-ish way of dealing with fixtures. In the end, it was also
possible to remove the MockSimParams class because it was not adding any
features on top of what SimParams can do already (and to what was
tested).
Change-Id: If5ef341e0585586127d5dae3f39dca2c232236f1
Signed-off-by: Jan Kundrát <jan.kundrat@telecominfraproject.com>
Getter and setter removed from the class PumpParams. The propagation
direction is cast to lower case string within the PumpParams
constructor.
Change-Id: Ice28affe8bcffbf8adcebb5cb096be8100081511
The error message was refering to method_nli which does not exist. The
correct parameter name is nli_method_name.
Change-Id: I24f24a2c5251317e1a80dda60aa27ec151628172
Signed-off-by: Jonas Mårtensson <jonas.martensson@ri.se>
The Raman solver gave the wrong loss profile when the fiber length was
not a multiple of the simulation parameter space_resolution as,
in this case, the fiber termination position was not included within
the considered z position array and the output loss profile was
evaluated at the wrong position.
Additionally, the Raman solver failed when the simulation parameter
space_resolution was greater than the fiber length as the z position
array contained only one element.
With this version the fiber termination position is always included
in the z position array and is composed of multiple uniformly spaced
positions and the final position (in general, the latter is not
uniformly spaced).
Example:
* fiber_length = 5, space_resolution = 4 => old version z = [0, 4]
=> new version z = [0, 4, 5]
* fiber_length = 5, space_resolution = 6 => old version z = [0]
=> new version z = [0, 5]
Alternative:
z = linespace(0, fiber_length, int(ceil(fiber_length/space_resolution)))
PROS
* Solves the first issue
* Returns a uniformly spaced z position array (there is not a
straightforward advantage in the simulation)
CONS
* Does not solve the second issue
* It is slightly more involved
Change-Id: I8886c3563ac7305c49cb5915712777ef561c5d4f
Bug: https://github.com/Telecominfraproject/oopt-gnpy/discussions/400
The actual conversion formula includes the minus (-), not the absolute
value. We never noticed it as GNPy simulates only modern networks
based on uncompensated transmission which have not DCUs. In this case,
the sign of beta2 along a path is the same for all the spans and,
in this case, the actual amount of NLI does not change.
Change-Id: I60a61d00c578a1a0436231a2bda8e3b6256fc8b3
I envision equipment.py as something which deals exclusively with the
traditional GNPy's JSON-formatted data, so make sure we do not include
any computation in that file.
Change-Id: I8473cccd84243147181a7195ba39fc6c9db3c42f
This mainly reverts some auto-fix-ups done in
I2f0fca5aa1314f9bb546a3e6dc712a42580cd562 which do not make that much
sense. By reverting them by hand, it's (hopefully) easy to see what is
just a tool work and what is an opinionated preference.
Change-Id: I6cb479e34b552fadc85c41b4b06b24e60c87b4a3
The Raman engine computes NLI just for a subset of channels; this is an
important speed optimization because the computation is rather CPU
heavy. However, the code just left NaNs in place for NLI of those
channels which were not explicitly simulated.
This is wrong because these NaNs propagate all the way to the total
input/output powers per the whole spectrum, etc, leading to NaNs being
shown to the user.
This patch uses a very simple linear approximation just in order to
prevent these NaNs.
fixes#288
Both of these places referred to "eq. 123 from arXiv:1209.0394", the
only difference (apart from the source of the input parameters, beta2
and asymptotic_length) was calling the two branches "SCI" and "XCI" vs.
"SPM" and "XPM".
In this commit I've only moved the code to a single implementation. The
input data are still being read from the same parameters, of course.
Conceptually, this is just about propagating the input parameters (which
drive the simulation) into all RamanFiber instances. The network module
already contains similar functions, let's move it there.
