AI-5G Introduction part -2

Home               LTE              NB-IoT          5G(NR-NSA)

How does AI integrate with 5G?

We all know that artificial intelligence is not only an interesting technology that improves accuracy and prediction on variety of problems but, it also  ultimately required to be used to take out the intelligence from the large amount of data produced on modern day networks. Also today’s data is not only large but it is growing so fast that about 80% of data is produced in last two years only.

Special industrial tenants are provided with a complete end to end virtual network by network slicing. We know that to ensure a good quality of experience (QoE) for industrial users in a virtual network is the key to successful network slicing. Now to improve the experience of the slice users it is primarily necessary to construct a wide range data map of the slicing. Data about slice users, Qos, events, subscription and logs can be collected in real time for multi dimensional analysis and then sliced into data cube. 

Using this data cubes artificial intelligence brain can be used to analyze, forecast and guarantee a healthy slicing. T the same time experience of multiple users can be used to analyze, evaluate and optimize to build user portraits to ensure a healthy, safe and efficient operation of slices.

In addition to all this the data cubes and AI brain together can serve for every process throughout the life cycle of slices, forming a closed loop. AI also helps to produce the slicing strategy and resolve slice faults in a smart way and optimize the performance automatically. 

This ultimately achieves smart scheduling of slice resources and give optimal configuration. This combination of AI and data cube will give effective guidance for smart life cycle operation of future slices.

We all know that 5G can offer virtually any service, but the importance of cognitive resource management cannot be underestimated. Also AI defined 5G radio access networks are proposed to support those unmatched and extraordinary requirements and leverage the emergence of mobile edge computing and caching, context aware networking and smart cities.

Also AI based 5G network provides the BSs (Base stations) or cloud with the capability to produce a cognitive and comprehensive data repository by splitting, processing and interpreting the operational data.

We all that, massive amounts of real time data are generated by a large number of users and this data can range from channel state information (CSI) to IoT device readings. This received data and its geo location database are fused to derive a complete understanding of the atmosphere. 

Thus from the perspective of the human centric communication, the human behaviors are learned and adapted by the AI defined networks to evolve the network functionalities and thus helps to create people oriented services. These AI defined networks are reconfigurable.

On the other hand from machine centric communication viewpoint, big data analytics are leveraged to extract massive patterns. Especially at physical and medium access controls layers and enable self organizing operations.

Also we can use neural networks to redefine communication networks. This can solve number of nontrivial design problems at runtime and across layers for cognitive link adaption, signal classification, resource scheduling and carrier sensing or collision detection among others.

The RNNS (Recurrent neural networks) is also capable enough to capture and mitigate the imperfections and nonlinearities of radio frequency components, like high power amplifiers which incur at physical layer and can affect the performance of network. A deep neural network (DBN) employs a hierarchical structure with multiple restricted Boltzmann machines and works through a successive learning process, layer by layer.

A CNN (Convolutional neural network) is built on layers of convolving trainable filters that result in a hierarchy of increasingly complex features.DBN and CNN are better suited for resolving a range of upper communication layer tasks like resource management and network optimization.

An artificial intelligence capability allows networks to identify problems such as service failures or a breakdown on the factory floor. This is then diagnosed and fixed automatically. Over the period of time it will also able to predict problems before they happen. Also AI can help telecommunication companies to design new 5G services by analyzing data in real time to ensure there are enough network resources and also point out where more resources are required.

Now most important is the matter of the edge. But it is? We all know that with the introduction of small and inexpensive processors more AI analysis and inference is going to take place not in the centralized cloud but at the level of smart phones and other devices.

This is known as a edge computing. It is basically a decentralized computing system that allows for data storage to be closer to the location where it is needed. These results in lower latency i.e. the time it requires for a request to travel from sender to the receiver and also for the receiver to process that request.This is true and required especially where large amount of data have to be processed immediately.

Lets us take an example of self driving cars. We know that these vehicles will need to make sense of large amounts of data from thousands of sensors like weather an object ahead is a person or debris and this has to be done continuously and in a matter of split seconds. But at the same time there is a lot of other data such as performance or predictive maintenance that can reside in a centralized cloud.

As some applications will require low latency, like streaming videos and others will not. For all this to perform smoothly network managers will need the ability to set priorities for traffic flows. Network slicing is seen as one solution. In network slicing single, shared physical networks has multiple virtualized networks running on top of it.

This will allow a manufacturer to pay for network slice with a guaranteed latency and reliability for connecting smart machines and equipment. Slices must be manually configured as far as current state of art is concerned. The AI can help with this as it can optimize the network so that traffic is routed based on device needs.

As said rightly every situation has two sides, so is with the relation of 5G and AI. This relationship of 5G and AI is not free of risks. Many think that the sheer amount of data that 5G may provide to AI could be obtained without any privacy. What has already happened in past when large amount of data  was maliciously extracted from social media by unauthorized organizations may occur again at a much larger scale with AI and 5G coupling.

---PINAL Dobariya



Post a Comment

Popular posts from this blog

5G Deployment Option-3/3a/3x

Authentication in LTE

Soft Handover vs Hard Handover

MAC-Contention Resolution Timer

DRX (Discontinuous Reception)

5G abbreviations

Carrier Aggregation & Dual Connectivity