Research proposal: Cell coverage prediction and optimization with machine learning

Introduction:

The expansion of mobile communication networks has led to the installation of a large number of mobile towers worldwide. The coverage area of a cell is a critical parameter for the network operator as it directly affects the quality of service (QoS) for mobile subscribers. To ensure high QoS, it is important to predict the coverage area of the mobile tower accurately and optimize it for maximum utilization with minimizing interference.

Inter-site interference has been always a limiting factor to achieving maximum efficiency of a cell. This interference is measured in terms of SINR and high SINR impact CQI (Channel Quality Index).

Before launching any new site, node parameters are designed in such a manner to best optimize cell coverage and quality and reduce interference. Further optimization is achieved through a process called DT (Drive Test), which is time taking and costly affair. Therefore, Iteration is mostly avoided. Despite all these plainings, there is always scope to further optimized coverage after the site launch.

Although with the introduction of new tools like geo maps and 3D maps. Location planning of site has become more accurate in term of traffic/load prediction; However, coverage prediction methods are still 20 years old and has not changed significantly with the evolution of technology.

An active network generates huge data. If we could leverage this available data and train a machine learning model on this data, we can predict the coverage of cells and tune it as per the best possible way.

Need for this study: Following gaps were identified in coverage planning and coverage optimization.

1. As mentioned in the introduction part, the current method of designing & planning cell coverage does not include a feedback loop to further check and evaluate cell coverage.

2. Current method of coverage measurement includes link-budget and drive test, which is very costly, time taking, and does not provide a feedback system.

3. Drive test method is majorly subjected to route availability and the skill of the DT engineer, this method covers a limited area and indoor is mostly.

4. TA (Timing Advance) analysis method consider the node as an isolated entity and TA analysis method do not consider the existence of a nearby node.

5. Most current method for coverage evaluation consider, the network as a static picture, however, the network behave as a living entity due to ongoing live changes such as

·       Addition of new sites and deletion of sites

·       Addition of new carrier (frequencies )

·       Network parameter changes such as (TX power)

·       New parameter and feature implementation

Objectives:

1.     The objectives of this research proposal are as follows:

2.     To develop an AI/ ML model that predicts the desired coverage area of a mobile tower.

3.     To develop an AI/ ML model that predicts overshooting of cell coverage.

4.     Further suggest optimize coverage

5.     To evaluate the accuracy of the AI model.

6.     To compare the performance of the proposed AI model with existing models.

7.     To investigate the impact of network event

Methodology: The proposed methodology for this research is as follows:

·       Data Collection: We will collect data from a network, that will include network topology and network configuration, the internal event information, and the corresponding design coverage area of each cell.

·       Data Preprocessing: We will preprocess the collected data by removing outliers and missing values.

·       Feature Engineering: We will engineer features from the collected data, including distance from nearby nodes, signal strength, and network load.

·       Feature selection: Using P-value statistics, we will filter significant features and drop the rest variables.

·       Model Development: We will develop an AI model, such as a neural network or decision tree, to predict the coverage area of a mobile tower.

·       Model Evaluation: We will evaluate the performance of the proposed AI model using metrics as per the model used.

·       Comparison with Existing design: We will compare the performance of the proposed AI model with the existing design.

Expected Outcome:

The expected outcome of this research is a robust AI model that predicts the coverage area of the cell. The proposed model will have higher accuracy with a combination of existing models, and we will demonstrate the impact of internal_event on the coverage area of the mobile tower.

This will help evaluate design coverage with actual coverage with minimum or almost no extra cost.

This introduces a feedback loop to estimate and optimize the best possible cell coverage.

This will also consider nearby nodes while estimating cell coverage.

This will help to reduce the overlap area between 2 cells and mitigate interference and improve QoS and traffic load.  

Conclusion:

The proposed research aims to develop an AI model that predicts the coverage area of a cell tower based on the handover with nearby nodes. This model will help mobile network operators to ensure high QoS for their subscribers by accurately predicting the coverage area of mobile towers. The proposed methodology includes data collection, preprocessing, feature engineering, model development, model evaluation, comparison with existing models, and investigation of the impact of handover with nearby nodes.

Note :

·       Special events (“which will use is input as fuel to estimate cell coverage“) are not mentioned, to protect the idea.

·       This is not technological advancement, its just an attempt to make use case of machine learning in the telecom optimization process.