You are on page 1of 4

Predicting Link Budget and Coverage area of B5G using machine

learning.

April 27, 2023


0.1. INTRODUCTION

By using machine learning techniques, a B5G coverage analysis method based on deep learning can be created to
forecast the coverage area and signal strength of the B5G network. This can be advantageous for network operators as
it allows them to plan their B5G network deployment and enhance the network’s performance. This article provides an
insight into machine learning techniques for the analysis with examples.

0.1 Introduction
Fifth-generation (5G) wireless networks offer faster data speeds and lower latency than their predecessors, but they also
have a shorter range and require more base stations and antennas for coverage. To optimize the deployment of 5G networks,
accurate coverage and signal strength predictions are crucial for network operators. Therefore, deep-learning-based methods
can be used to predict the coverage area and signal strength of the B5G network using a large dataset of simulated or
measured B5G coverage data to train a deep learning model. This approach has the potential to help network operators
plan and optimize their B5G network deployments.
Previous research has highlighted the potential of deep learning for 5G network planning and optimization, as discussed
by Chen et al. (2021) and Li et al. (2020). Moreover, Vakilian et al. (2021) proposed a machine learning-based approach
for predicting B5G network coverage. These studies demonstrate the utility and growing interest in deep learning-based
methods for predicting coverage and signal strength in 5G and B5G networks.

0.1.1 Why do we need machine learning for predicting


Machine learning can be useful in predicting link budget and coverage area for beyond 5G communication for several
reasons:
Complexity: Beyond 5G communication systems are expected to be more complex than previous generations, in-
corporating a range of new technologies such as massive MIMO, beamforming, and network slicing. Machine learning
algorithms can help manage this complexity by automating the analysis and prediction of these systems.
Variability: In addition to increased complexity, beyond 5G communication systems are expected to be highly variable
due to factors such as environmental conditions, user behavior, and interference. Machine learning algorithms can help
account for this variability by learning from historical data and adapting to changing conditions.
Efficiency: Traditional methods of predicting link budget and coverage area typically require significant manual effort
and computational resources. Machine learning algorithms can help streamline this process by automating many of the
steps involved and reducing the computational burden.
Accuracy: Machine learning algorithms can improve the accuracy of link budget and coverage area predictions by
leveraging large datasets and incorporating a range of factors that may impact performance. This can help ensure that
beyond 5G communication systems are designed and deployed with optimal performance and reliability.

0.2 Methodology
The B5G coverage analysis method based on deep learning involves the following steps:
Data Collection: A large dataset of simulated or measured B5G coverage data is collected. This dataset includes
features like signal strength, location, terrain data, and other relevant variables.
Data Preprocessing: Input features are preprocessed to normalize and transform them into a suitable format for the
deep learning model.
Feature Extraction: Relevant features are extracted from the input data using handcrafted and learned techniques.
Deep Learning Model: A deep learning model, such as a convolutional neural network or recurrent neural network, is
trained on the preprocessed and extracted features to predict coverage area and signal strength for a given input.
The deep learning model can be represented mathematically as a function that maps input features to predicted coverage
and signal strength outputs. Let x be the input feature vector that represents a specific location in the network, and y be the
output vector that represents the predicted coverage area and signal strength. The deep learning model can be represented
mathematically as a function that maps input features to predicted coverage and signal strength outputs. Let x be the input
feature vector that represents a specific location in the network, and y be the output vector that represents the predicted
coverage area and signal strength. The deep learning model can be represented as follows:
y = f (x; θ )
where θ represents the model parameters that are learned during training and f is a nonlinear function that maps the
input feature vector to the output vector. The deep learning model is trained using a loss function that measures the
difference between the predicted output and the true output. The loss function can be represented as follows:
L(θ ) = ∑ ||y − f (x; θ )||2

2
0.2. METHODOLOGY

where |||| represents the L2 norm. During training, the model parameters θ are updated using backpropagation and
stochastic gradient descent to minimize the loss function. Once the model is trained, it can be used to predict the coverage
area and signal strength of new locations in the network.
Postprocessing: The output values generated by the deep learning model are post-processed using techniques like
smoothing and filtering to remove noise and improve prediction accuracy.
Feature Fusion (Optional): Multiple sources of input data, such as satellite imagery and weather data, can be combined
using the feature fusion step to enhance the accuracy of coverage predictions.
Data Collection:
B5G coverage data can be collected using simulation or measurement techniques. Simulation provides a large dataset
of coverage data, while measurement provides more accurate data but with a smaller dataset. The dataset should include
features such as signal strength, location, terrain data, and other relevant variables.
Data Preprocessing:
Input features are preprocessed to normalize and transform them into a suitable format for the deep learning model.
Common preprocessing techniques include normalization, scaling, and feature selection based on relevance to the prediction
task.
Feature Extraction
Handcrafted and learned techniques are used to extract relevant features from the input data. Handcrafted features are
designed by experts based on their domain knowledge, while learned features are generated automatically by the deep
learning model.
Deep Learning Model:
A deep learning model is trained on the preprocessed and extracted features to predict coverage area and signal strength
for a given input. The model is trained using a loss function that measures the difference between the predicted output and
the true output.
Postprocessing:
The output values generated by the deep learning model are post-processed using techniques like smoothing and
filtering to improve prediction accuracy.
Feature Fusion:
The feature fusion step can be used to combine multiple sources of input data, such as satellite imagery and weather
data, to enhance the accuracy of coverage predictions.

Figure 1: Methodology Steps

3
0.3. EXAMPLE WITH CODE

0.3 Example with Code


Let’s take a fictitious scenario to show how the deep-learning-based B5G coverage analysis approach functions. Let’s say
we want to set up a 5G network in a city with a lot of tall structures. We’ve compiled a dataset of measured B5G coverage
information for this region, which includes the locations of 5G base stations, the signal strength at various locations, and
the structural details of the buildings. With the use of this dataset, we may forecast the 5G network’s coverage area and
signal intensity using a deep learning model, such as a convolutional neural network (CNN). An region map with the
positions of existing base stations and the physical details of the buildings serves as the model’s input. With the use of
this dataset, we may forecast the 5G network’s coverage area and signal intensity using a deep learning model, such as a
convolutional neural network (CNN). An region map with the positions of existing base stations and the physical details
of the buildings serves as the model’s input. A map displaying the anticipated 5G network coverage region and signal
intensity is the model’s output.
Once the model has been trained, we may utilize it to optimize how the 5G network is deployed. The model can be used
to simulate various situations, such as the addition or relocation of base stations, and to forecast the network’s coverage
area and signal strength in each one. The option that offers users the best coverage and signal strength can then be selected.

0.3.1 Code
1 from keras . models import Sequential
2 from keras . layers import Conv2D , MaxPooling2D , Dense , Flatten
3
4 model = Sequential ()
5 model . add ( Conv2D (32 , kernel_size =(3 , 3) , activation = ’ relu ’ , input_shape = input_shape ) )
6 model . add ( MaxPooling2D ( pool_size =(2 , 2) ) )
7 model . add ( Conv2D (64 , kernel_size =(3 , 3) , activation = ’ relu ’) )
8 model . add ( MaxPooling2D ( pool_size =(2 , 2) ) )
9 model . add ( Flatten () )
10 model . add ( Dense (128 , activation = ’ relu ’) )
11 model . add ( Dense (1 , activation = ’ linear ’) )
12
13 model . compile ( loss = ’ m e a n _ s q u a r e d _ e r r o r ’ , optimizer = ’ adam ’)
14
15 history = model . fit ( x_train , y_train , batch_size =32 , epochs =10 , v al id a ti on _d a ta =( x_val , y_val )
)
16
17 test_loss = model . evaluate ( x_test , y_test )

From line 1 to line 2 , required libraries are being included. Here we are using Keras library which is a highlevel
machine learning library based on Tensorflow. A CNN model has been defined in the lines from 4 to 11. On line 15
training is being done. At end we can evaluate the model to check how accurate the model is.

0.4 Conclusion
For forecasting the coverage area and signal intensity of B5G networks, the deep-learning-based B5G coverage analysis
approach is a potential tool. The technique trains a deep learning model that can forecast the coverage area and signal
strength for a given input using a sizable dataset of actual or simulated B5G coverage data. The technique can be used to
streamline the rollout of 5G networks and guarantee that the network offers users enough coverage and signal strength. The
deep-learning-based B5G coverage analysis method has the potential to completely alter how 5G networks are developed
and implemented with additional research and development.

You might also like