Using tuning, you can calibrate the Synapse model. The aim is to reduce the effects of imperfections and to enable the Synapse model to be as closed to real propagation measurements as possible.

Description

A fast and automated tuning module is integrated into the Synapse model that takes into account drive tests data and, as a result, improves prediction data and generates results that are closer to the real world.

The tuning process results in the generation of two analyses that are used to adjust the coefficients of the model. In the last step of the tuning process, the Synapse model generates new predictions using the drive tests points with the tuned model.

The tuning algorithm implemented in the Synapse model is based on a multi-linear regression method. The model recognizes that the amount of total loss is partly dependent on statistically estimated losses. This adjustable part is a sum of variables multiplied by coefficients. The tuning algorithm determines the value of these coefficients with the best fit of the test measurements. Because the statistical adjustment has no physical meaning, the Synapse model combines elementary physical theories prior to statically adjusting results. As the statistics are not the bases of the prediction results, poor measurements will not result in an unreliable tuned model.

Tuning process

During the tuning process, the calibration tool optimizes coefficients for each type of morphology. Note that if the morphology option is not selected, the model will only use the default morphology meaning a unique optimization for all environments. Before starting the tuning process, the calibration tool will separate the measurement points into different categories (boxes) according to the morphology they belong to and according to other criteria (typology depends on the geographical data available on the considered area, Line of Sight, frequency, micro cellular context, etc.).

Note that there is an initial checking based on the minimum number of points: a minimum of 1000 measurements points per box is required to start optimization otherwise the model uses default parameters.

The calibration process allows you to generate two different calibrations (it means two different sets of coefficients):

•The first one, which will be apply when polygons are available on the calculation area.

•The second one, which will be apply when polygons are not available on the calculation area.

The coefficients, which are calculated by the smallest square method, are passed as input parameters to the new Synapse tuned model.

The calibration tool performs a specific calibration for each box but it first checks the distribution of each variable calculated along the profile; it means that the calibration tool will associate the variables calculated along the profile with the best coefficients to reduce the average and the standard deviation but only the variables considered as representative of the real-world environment. The other variables will be associated to a default coefficient (ʹrepresentative of the realityʹ means in adequacy with what the model learnt during all steps of its development).

Note: The calibration tool optimizes certain parameters (provided enough points are available), which means the model can show very good statistics, depending on the measurements used. Results can be a little less favorable than with the other option but they will be more robust when using the same model in other type of environment.

Warning: The transmitters and the receivers located inside a building are not considered during the Synapse Model's Tuning process.

How to tune the Synapse model in Xirio

In order to launch the tuning process, you previously need measurements in your multitransmitter coverage study. These measurementes must be uploaded in a file (with the appropriate measurements format) using the option "Calculation and visualization of route" on the "Calculate study" section.

Then, also on the "Calculate study" section click on "Calculate adjustment of measures":

In "Method to calibrate" section you have the different options for tuning Synapse model:

•When you select the "Use automatic filtering" checkbox, the tuning engine filters out drive tests where inconsistent values have been found, it means drive tests for which the mean error before tuning is outside the interval [-10;10] (these drive tests can generate inconsistency in the tuning coefficients). Note: When this option is selected, the process calculates and displays the percentage of measurement files for which the mean error is comprised in the interval [-7;7] before the calibration (for example, if 10 measurement files are available, and for one the mean error is outside the interval [-7;7], the quality will be 90%), a value close to 100 indicates that the drive tests files are of a good quality. This information is called Measures quality indicator and is displayed in the model tuning log.

•When you select the "Do not consider indoor pixels for raster areas" checkbox, the tuning engine filters out drive tests which are located inside Building clutter classes. When it is not enabled, drive tests which are located inside Building clutter classes are not considered as indoor, it means that tuning engine automatically sets the clutter classes type Building to Other.