# Deep-Mobility: A Deep Learning Approach for an Efficient and Reliable 5G Handover
###### tags: `5G Reading`
Date : 2022-10-14

## Metadata
[paper link](https://arxiv.org/ftp/arxiv/papers/2101/2101.06558.pdf)
Paropkari, R. A., Thantharate, A., & Beard, C. (2022, March). Deep-Mobility: A Deep Learning Approach for an Efficient and Reliable 5G Handover. In 2022 International Conference on Wireless Communications Signal Processing and Networking (WiSPNET) (pp. 244-250). IEEE.

## Take away
**Why Handover is important in 5G**
Due to the **high frequency** characteristic of 5G, small cells have a very important role in providing 5G connectivity to the end users. With the exponential increases in devices, data and network demands make it mandatory for the service providers to manage handovers better, to cater to the services that a user desire.

**How does the study done their handover**
The proposed DEEPMOBILITY model collects the data about the UEs, RF, and the network, and **predict the HO decision** through a deep learning model with 4 hidden layers. 30% of the data are used for validation.

## Summary
With the recent advancements, the management of cellular handovers has become more complex. In addition, RF and channel conditions may change due to several environmental and surrounding factors. Human intervention at every critical step isn’t enough, so SONs are gaining importance.
This study utilizes a Deep Learning model to help making the HO decisions. By collecting the related parameters; such as network key performance indicators (KPIs) and RF signal conditions, and feed into the models built from Recurrent neural network (RNN) or Long Short-Term Memory network (LSTM) for different handover goal. The proposed model is also straightforward and flexible; therefore, it will be useful for formulating the 5G/6G handover algorithms that will need to incorporate hyper-network densification, mmWave, M2M traffic, and ultra-reliable low latency communication also.

## Note
- The Huge demands on data and bandwidth
According to Ericsson’s Mobility report from November 2018, the global mobile data traffic is expected to increase eight times between 2017–2023 and video content about five times being offered on over 70% of whole mobile data traffic triggering the **eMBB(enhanced mobile broadband)** services.
Also, the service providers are also forced to provide an infrastructure for
serving various applications and business with superior Quality of Service (QoS) for some special use case; such as public safety, emergency and medical, in **URLLC(ultra-reliable low latency communication)** aspect.
In the IoT(internet of things) domain, there are another pool of devices that requires the **mMTC(massive machine type communication)** services of the 5G networks transforming the home, office, city streets, public places, and beyond for their requirements on Quality of Service (QoS).

- Important factors in making Handover decision
Network mobility was one of the prime factors to trigger a handover in any given network, but several other parameters like **Received Signal Strength (RSSI)**, **Signal to Interference and Noise Ratio (SINR)**, **Fade Duration (FD)**,**Quality of Service (QoS)**, **backhaul connectivity**, **network congestion**, **reliability**, etc., may also result in a handover.

- Some key points for the Handovers
- There are **hard HO**(Handover) procedures in LTE which defined in 3GPP LTE-Advanced spec.
- hard HO requires **Handover Margin (HOM)**, **Time-To-Trigger (TTT)** and **A3offset**. A3event would be the initialized condition for hard HO. RSRP (signal power) and/or the RSRQ (signal quality) as a single parameter considered to trigger a handover.
- The Handovers need to keep up a balance in order to **avoid the ping pong effect** and also **too much loading of any one** specific cell.
- Some of the SON functions are designed to handle the mobility, such as Mobility Robustness Optimization(**MRO**) and Automatic Neighbor Relation(**ANR**). When doing the optimization, avoiding the **parameter confliction** are also need to be considered.
- HO decision can be made using multiple other parameters like the bandwidth capacity of a gNB, current user load, backhaul capacity, future maintenance activities in the database, RF channel stability, modulation and coding schemes (MCS), dual connectivity (DC), etc.

- The DEEPMOBILITY Model
- Collected data:
- UE type used
- UE supported technology
- Time and day of the connection, RSRP, RSRQ of serving as well as some (3-4) neighboring cell sites
- RF channel conditions based on MCS
- Available channel bandwidth
- Backhaul capacity
- Alarm status on cell sites
- Maintenance tickets if any, etc
- What the model does:
- Too many HOs and when/where neighbor list needs to be updated
- Power levels of each eNB at every location, and in UL
- RF conditions
- Applications being used and other service parameters the from network and RF perspective.
- Behavior of the UE:
- Movement and velocity of the UE
- Routine mobility patterns of the UE
- UE bandwidth demands
- Model structure
1 input layer, 4 hidden layers and 1 output layer. Each layer is activated using a certain activation function to make parameters non-linear and close to the real world. A 4x3 means a total of 12 neurons in that particular layer. Output layer is applied a simple linear regression function.