LEARNING-BASED METHOD AND SYSTEM FOR CONFIGURING AN OPTICAL TIME-DOMAIN REFLECTOMETER IN A GIGABIT PASSIVE OPTICAL NETWORK
20220376782 · 2022-11-24
Inventors
Cpc classification
H04B10/0773
ELECTRICITY
International classification
Abstract
The present disclosure provides a method and system for configuring an optical time domain reflectometer (OTDR) in a gigabit passive optical networks (PON), characterized by the steps of: collecting network data of the network to be scanned by switch controller to characterize said network; collecting data from various optical network terminals (ONTs) of the gigabit passive optical networks (GPON) by an OTDR and the Switch Controller to form a training database, the training data is used to train the method; optimizing the parameters of the optical time domain reflectometer (OTDR) based on the network data and the training database by a processor provided on the switching controller using machine learning. For point-to-multipoint (PMP) networks such as PON, the present method and system provides optimal set of parameters to configure OTDR before performing trace.
Claims
1. A method for configuring an optical time domain reflectometer (OTDR) in a gigabit passive optical network (PON), comprising: collecting network data of the network to be scanned by a switch controller to characterize said network; collecting data from various optical network terminals (ONTs) of the gigabit passive optical network (GPON) by an OTDR and the switch controller to form a training database, the training data being used to train the method; and optimizing the parameters of the optical time domain reflectometer (OTDR) based on the network data and the training database by a processor provided on the switch controller using machine learning.
2. The method as claimed in claim 1, wherein the parameters of optical time domain reflectometer (OTDR) are selected from the group consisting of pulse width, acquisition time and distance range or a combination thereof.
3. The method as claimed in claim 1, wherein the network data is selected from the group consisting of maximum distance of the fibre from optical line terminal (OLT) in the GPON, link loss and optical return loss or a combination thereof.
4. A system for configuring an optical time domain reflectometer (OTDR) in a gigabit passive optical network (GPON), the system comprising: a switch controller configured for collecting network data from a network to be scanned characterizing said network, the switch controller having: an optical time domain reflectometer (OTDR) to be configured, the OTDR and the switching controller operable for collecting data from various optical network terminals (ONTs) of the gigabit passive optical network (GPON) to form a training database; an optical switch; and a processor adapted to configure the optical time domain reflectometer (OTDR) by optimizing the parameters of optical time domain reflectometer (OTDR) based on the network data using the training database.
5. The system as claimed in claim 4, wherein the optical switch is configured to receive signal from optical time domain reflectometer (OTDR) while scanning said network to be scanned.
6. The system as claimed in claim 4, wherein the switch controller further comprises a plurality of wavelength division multiplexing (WDM) coupler to couple the output of the optical switch to the gigabit passive optical networks (GPON).
7. The system as claimed in claim 4, wherein the system further comprises a plurality of passive power splitters (PS) to split coupled output received from WDM coupler towards various ONTs.
Description
BRIEF DESCRIPTION OF DRAWING
[0021] The above and other features, aspects, and advantages of the subject matter will be better understood with regard to the following description, and accompanying drawings where:
[0022]
DETAILED DESCRIPTION OF THE INVENTION
[0023] An exemplary embodiment of the device as indicated in
[0024]
[0025] Our proposed method runs in a card of switch controller (120), which is integrated to GPON solution as depicted in the
[0026] The switch controller (120) will take 16 PON inputs from OLT (110), and one input pulse from OTDR (121) embedded on the switch controller (120). WDM coupler (124) of switch controller (120) sends the multiplexed signal towards the PON side after coupling the signals from OLT (110) with OTDR pulse. All PON networks are independent and may have any number of ONTs (upto 128) and other components. The system may have various passive optical splitters (PS) (130) to split the multiplexed signal received from coupler towards various ONTs.
[0027] Whenever OLT detects no upstream power i.e., from ONT to OLT, it is declared as loss of signal (LOS). It is immediately sent to switch controller to acquire fault trace via control path. Switch controller, based on the PON number associated with the fault, selects the appropriated port of optical switch and triggers the OTDR to acquire the trace. Similarly, it switches the port and takes trace if any other fault in different PON is detected. So far OTDR parameters are configured only once by the user interface (UI) and remain same till it is changed again. If, for example, a particular pulse width, which decides the power injected into the fiber and thus decides the distance it can travel, is selected, it might be good enough to see one complete PON but it might not be able to see even half of the network for another PON. Similarly, other parameters may affect the accuracy and characterization of PON. Now the present disclosure provides a method to select optimum parameters for a particular network or PON based on the network configuration, before taking trace. Algorithm for predicting parameters is trained using machine learning approach.
[0028] Regression algorithms, a supervised machine learning approach are used in the present disclosure. Following is the brief introduction of the algorithm. Regression algorithms belong to family of Supervised Machine Learning algorithms. Purpose of supervised learning algorithms is to model the dependencies and relationships between the output and input features or dependent and independent variables, to predict the value for new data. The algorithm builds a model on the features of training data and using the model to predict value for new data. The simple linear regression attempts to establish a linear relationship between one independent variables and a dependent variable. In multiple linear regression model there are two or more independent variables and a dependent variable. Whereas in multivariate multiple linear regression both independent variables and dependent variable are two or more.
[0029] As there are three independent and three dependent variables i.e, there is a need to choose three OTDR parameters based on three network attributes, so multivariate multiple linear regression method is used to establish the relationship. The present disclosure is providing a brief introduction to multiple linear regression. The general model for multiple linear regression with k independent variables is of the form
y.sub.i=β.sub.0+β.sub.1x.sub.i1+β.sub.2x.sub.i2+ . . . +β.sub.kx.sub.ik+.sub.i, i=1,2, . . . ,n.
[0030] There are total n observations and above equation signifies ith observation, where y, is dependent variable, x=[x.sub.i1, x.sub.i2, x.sub.i3, . . . , x.sub.ik] are the k independent variables, is the estimation or prediction error and β=[β.sub.0, β.sub.1, . . . , β.sub.k] is a vector of regression coefficients. To simplify the computation, we have written the multiple regression model in terms of the observations using matrix notation. Using matrices allows for a more compact framework in terms of vectors representing the independent variable, dependent variables, regression coefficients, and estimation or prediction errors. The model takes the following form
Y=Xβ+
and when written in matrix notation, we have
[0031] It can be noted that Y is an n×1 dimensional random vector consisting of the observations, X is an n×(k×1) matrix determined by the predictors, β is a (k×1)×1 vector of unknown parameters, and is an n×1 vector of random errors.
[0032] The first step in multiple linear regression analysis is to determine, using training data, the vector {circumflex over (β)}, which gives the linear combination ŷ that minimizes the length of the prediction error vector. In other words, the vector {circumflex over (β)} minimizes the sum of the squares difference between ŷ and y and later on this vector is used to predict dependent variable y.sub.i when any new test data come. Now, since the objective of multiple regression is to minimize the sum of the squared errors, the regression coefficients that meet this condition are determined by solving the least squares normal equation.
X.sup.TX{circumflex over (β)}=X.sup.TY
[0033] An important assumption in multiple regression analysis is that the variables x.sub.1, x.sub.2, . . . , x.sub.n be linearly independent. Now if the variables x.sub.1, x.sub.2, . . . , x.sub.n are linearly independent, then the inverse of X.sup.TX will exist, and we can obtain
{circumflex over (β)}=(X.sup.TX).sup.−1X.sup.TY
[0034] Similarly, regression coefficients for other dependent variables can be estimated.
[0035] Following are the description of OTDR parameters and their impact on characterization of network in the form of trace generated by OTDR. They have been chosen as dependent variables or output variables. [0036] 1. Pulse Width: Pulse width is the most important OTDR parameter as it affects the dead zone. The OTDR dead zone refers to the distance (or time) where the OTDR cannot detect or precisely localize any event or artifact on the fibre link. If two events fall in the dead zone, OTDR cannot make distinction between two and treat them as one event. It creates severe challenges and need to be solved by using appropriate pulse width. Narrow pulse widths can see more detail on the link and can identify events that are closer, but also produce noise due to low signal-to-noise ratio (SNR). Longer pulses allow OTDR to span longer distance but with higher dead zone. [0037] 2. Averaging Time or Acquisition Time: Averaging time decides the number of measurements averaged together to create a trace. This can vary from a few seconds to three minutes. A short averaging time decreases testing time but results in noisy traces, while choosing longer averaging time increases dynamic range and accuracy. Longer averaging time tends to cancel noise in the waveform and produce smooth trace. [0038] 3. Distance Range: Distance Range defines the maximum distance from which the OTDR can acquire data samples. The longer the range, the further the OTDR will shoot pulses down the fibre. If the range is set incorrectly, the trace waveform may contain undesirable artifacts, such as ghosts.
[0039] Following are the PON network attributes that have been chosen as independent variables or input variables. [0040] 1. Maximum Distance: It is the maximum distance of the fibre from OLT in the network. It is important parameter to choose pulse width and range. [0041] 2. Link Loss: It is the total loss in the network, due to fibre attenuation, splitter loss, connector loss etc. It is important parameter to choose pulse width. [0042] 3. Optical Return Loss (ORL): It represents the total reflected optical power from a complete fibre link, includes the portion from backscattering as well as the reflected power from optical connectors and medium discontinuities.
[0043] Following are the various steps involved in the method. [0044] 1. First step of any supervised learning is to collect data for training. For that networks with different topology have been created i.e, different level of split, different maximum fibre length, different number of ONTs etc. Then OTDR parameters are configured manually to characterize the network accurately i.e we have chosen pulse width, range and time which give us best characterization of network. In this way we have created labelled data for each network. Three independent and three dependent variables per network. Similarly, this kind of data is generated for each network. [0045] 2. Second step is to use data that we have generated in step 1, to find the regression coefficients (unknown variables in regression model) using the model described above. [0046] 3. Once the coefficients have been found, any test data can be performed and to predict the output which is optimum in least square sense.
[0047] Following are the different steps involved. [0048] a. Acquire trace at pulse width (3 us) and Range (60 km) as these are good enough to see GPON. It can be changed depending on fibre networks. [0049] b. Resulting trace will give us maximum distance of the fibre under test, link loss of the network and ORL of the network. [0050] c. Now put these three input or independent parameters in regression equation to find three dependent parameters using the model and regression coefficients found step.