How AI will shape radio network design jobs and tools in near future

[henry3499]

In past few years Artificial Intelligence become an important gear in many fields for solving performance problems and use of it's powerful predictive model in business marketing, politics, weather forecasting and modern technologies.
In this article, will focus on the potential applications and use case of Artificial Intelligence (AI) in Radio Network Design (RND) tasks and it's consequences on the future of exiting planning tools (Atoll, Planet, Asset,...) and RND Jobs.
Artificial Intelligence is a large term but if we simplify its applications to Radio Network Design, we can define it as the use of machine learning algorithms to create a predictive model. The predictive model is trained using a predefined data-set of inputs and outputs and each input entry will correspond to an output entry. Once the Model is trained, it will be possible to provide different inputs and the model will predict the corresponding future outputs. The model could apply a simple linear equation or use complex neural network methods.
The Radio Network Design is a process performed by engineer to build a green field network or to identify new coverage and capacity expansion on existing network.
The main deliveries of Radio Network Design consist of following:

1. List of sites location with latitude and longitude to fulfill the coverage and capacity requirements.
2. Required technologies (GSM, UMTS and LTE) per sites and the number of layer required by technology (GSM900, GSM1800, UMTS900, UMTS2100, LTE1800,..)
3. Radio parameters configuration and are mainly : Azimuth, Electrical and Mechanical Tilt, Antenna height, Transmission Power
4. Additional Capacity solution like Macro and Micro Small cells, Massive MIMO,...

We will discuss in this article the use case of Artificial intelligence in above Radio Network Design deliverable.

• List of sites location with latitude and longitude to fulfill the coverage and capacity requirements.

As the machine learning algorithm are only as good as their datasets availability, it will be difficult to provide different site location inputs and it's correspond performance. In one particular geographical area, it will be not a practical scenario to put a site and then remove it in order to create a datasets for training the model for site selection.
The identification of site location is based on propagation model which are used to predict the signal level at each pixel of the area. Most of the existing propagation models are empirical models and require a model calibration using Continues Wave (CW) measurement and purchase of geographical maps (clutter and elevation). Propagation model calibration is a time taking and costly process.
The potential scope of using Artificial intelligence (AI) is the creation of a predictive propagation models using neural network. Drive test or live measurements reports (Call traces,..) can be used to train the model since for each location (X,Y) we have a correspondent coverage level. Once the model is trained, it can be implemented to predict the signal level at each pixel of the area.
Below figure is showing my realized propagation model based on neural network. Drive test measurements are used to train the model and Levenberg Marquardt method is used.

The neural propagation model can be used to run a site selection on existing sites or to identify the coverage hole to place new sites.
It is possible to replace the traditional models with Neural model to run predictions plot.

• Required technologies (GSM, UMTS and LTE) per sites and layer requirements by technology (GSM900, GSM1800, UMTS900, UMTS2100, LTE1800,..).

Capacity in mobile network is the amount of Circuit Voice (CS) and Packet Circuit (PS) generated by the users to be carried by the designed network. For deployment of new technologies like 5G, we can use the existing traffic of LTE multiplied by a growth factor. same is applicable for LTE by using UMTS network.
The capacity dimensioning is a process to determine the number of base station and the required layers and bands.[/COLOR]
The forecast factor calculation for the future number of users by using the Artificial intelligence is a promoting method as it will offer a cell wise capacity dimensioning. The drawback of this method is it will require the use of large historical traffic dataset of an existing technology to train the predictive model.
Also by having an historical data for the sites and layer augmentation during past years in the network, trained neural model could predict the required sites and layer expansion but still it is recommend to use the traditional guideline for the capacity dimensioning for carrier addition.

• Radio parameters configuration and are mainly : Azimuth, Electrical and Mechanical Tilt, Antenna height, Transmission Power.

The existing Automatic Cell Planning (ACP) module are still required to tune the above radio parameters using traditional algorithm and the use of Artificial Intelligence (AI) will be limited to network tuning and optimization using Self Organized Network (SON) software for Coverage and capacity load balancing (CCLB).
Self Organized Network (SON) can be used to create the model training datasets by changing the capacity load balancing (CCLB) parameters and at the same time measure the corresponding network performance.
Once we have different network performance for different parameters values, it will be possible to propose the optimum optimized parameter values using the trained neural model implemented inside the SON.

• Additional Capacity solution like Macro and Micro Small cells, Massive MIMO,...

Combination of data analytics and artificial intelligence is a future option for hot spot identification to place a small cell. Google map measurements data, YouTube, Facebook, Whats-app and other social media data inputs can be used as training datasets by machine learning algorithm. The output of the predictive model is the potential locations where we we can place an indoor solution.
Below figure is showing an example of capacity hole map and proposal of small cell solution.

Conclusion
The scope of Artificial Intelligence in radio network design is limited to certain design task and competences in link budget calculation and radio propagation will be still required.[/COLOR]
Artificial Intelligence will majorly affect the network optimization in large scale due to availability of live measurements which are used to develop the AI predictive models . Existing optimization skills need to be upgraded to next level in machine learning and data analytics!
Artificial Intelligence features development and implementation in planning tools can be a good future road map to improve the tool sustainability.
Thank you for reading and please share