Towards 5G, Artificial neural networks and EDGE computing Part-1

The digital disruption will be the new age digital revolution similar to the 1700s or early 1800s industrial revolution and its impact on modern society. Mobile data traffic is rising rapidly due to new paradigms such as virtual reality, augmented reality, mixed reality, autonomous vehicles, V2X, real time multi-player gaming, massive IoT like billions of connected intelligent IoT devices, and sensors. With billions of connected devices, the end user has exponentially growing number of connections. So, existing network technologies are fast becoming inept at handling these kinds of new communication wave which need improved energy efficiency models, with ultra-low latency, and high bandwidth requirements.

So, in this blog, we will discuss the 5th generation network system, EDGE computing, artificial neural network-based machine learning models and its application on edge systems and 5th generation communication networks.

EDGE computing is processing the data near to the network where data is being generated.

The global mobile data traffic has grown by 69% in 2018 and is expected to increase another seven-fold in the near future. Virtual reality video streaming needs high throughput (~200Mbps) and low latency (~10ms). Autonomous vehicles are connected to 5G networks all the time and there is a need to process more transportation, guidance data than existing vehicle data usage. So, a future system with low latency, high bandwidth, and advanced data analytics need to fill this gap. Here, the combination of 5G, EDGE, and deep artificial neural network-based machine learning models can be useful.

Brief introduction to the key components of our blog.

5^th Generation Communication:

5G network system is a much needed, evolutionary technology that will be used to improve future massive machine type communication, AV, V2X, M2X, etc. It provides endless possibilities for service providers, network operators, and consumers with a different set of QoS, dynamic changing business requirement and improved solution.

5G communication network promises increased capacity, ultra-high network bandwidth (~ 20 Gbps or more) and ultra-low latency communication (< 1ms). For example, manufacturing industries have a number of different, time critical IoT and associated sensor devices. Existing wireless technologies are not developed for handling the ever-increasing IoT and sensor type communication. But the future of the industrial domain will heavily depend on the billions of sensors based IoT devices. So, it needs better handling of these devices with intelligent behavior. 5th generation technology uses new air interface called as New Radio (NR), which uses existing 600 MHz-6 GHz and millimeter wave bands of 24 GHz-86 GHz.

Some of the key features of 5G network are listed below,

SDN (Software Defined Networks) and NFV (Network Functions Virtualization)
Massive MIMO
Ultra-reliable low latency connection
Small Cells
Network Slicing mechanism
Massive connected IoT devices
Millimeter wave

5G system includes the following components,

eMBB (enhanced Mobile Broadband)
URLLC (Ultra Reliable Low Latency Communication)
mMTC (massive Machine Type Communication)

For more details, please refer to the References section at the end of this blog.

Network Slicing:

The world is moving toward an advanced digital age where more and more digital applications, services, and devices are involved, bringing in more data transformation with specific use case capabilities. Network slicing provides multiple logical networks to be created over a common shared physical infrastructure. Network slicing is the combination of two well-known network virtualization technologies such as software defined networking (SDN) and network function virtualization (NFV). In this, the software based logical partition is crucial to provide dynamic allocation of physical resources based on the different service requirements. 5G network slicing has intelligent machine learning based orchestration.

EDGE Computing:

5G network depends on low latency communication and process of massive connected devices data. It is imperative to use edge computing for this purpose. EDGE computing processes the data near the network where it is generated. It reduces the centralized cloud-based data processing. It provides organizations with the capability to access data in real-time, process it with low latency requirement, and deliver crucial suggestion and prediction in quick time. This will help different 5G applications such as,

Low latency industry applications
Autonomous cars
Public safety applications and agencies
High bandwidth and low latency-based AR/VR/MR applications,
Better secure communication for critical applications with less processing nodes

EDGE computing is also known as Mobile EDGE Computing (MEC). Also fog computing is the standard of edge computing used in some places.

For example, multiple high-end security camera applications are used to send real-time images and videos to the cloud for intelligent processing. This will make the end cloud network assign more resources for that specific application and more bandwidth and data in the intermediate network, which will create network congestion. Instead of this, if we process the priority/ criticality set of security camera application data at a nearby edge device and intelligently identify which data needs to be processed at cloud, it will improve the end- to- end communication.

Overall, edge computing will be more beneficial to the next generation computing and billions of connected IoT elements. Now will just processing the data and minimizing workload of cloud network be enough for the next generation communication system? The answer is NO. If massive device data with hybrid network types and different network slices are involved, then definitely, a very intelligent processing power is needed. Here, deep neural network and machine learning techniques can come to the fore.

Artificial Neural Network (ANN):

Neural networks are a specific set of machine learning algorithms which resemble the human brain and are made up of artificial neurons. ANN will take multiple inputs, which pass through multiple hidden layers and multiple feedback loops, and finally produce a more accurate single output. ANN will find out complex patterns in Big Data that would otherwise be impossible to detect easily.

Now a days, deep neural networks have made a huge impact on a large number of complex tasks in AI (Artificial Intelligence) field and its wide area of applications. But, the impact of neural algorithms on EDGE devices such as telecom devices, gateways, consumer devices, vehicles, and avionics devices are limited and have a big challenge to implement with high accuracy. But it has a huge potential to improve the performance and security of EDGE based 5th generation telecom network and systems.

For example, retail and banking are the most vulnerable targets for cyberattacks. Any compromise in the customer data could lead to a loss of brand trust, business, and market. Different domains such as public safety, defense, aerospace, and organization need better mechanisms to defend the existing as well as unseen problems.

Processing huge data of connected IoT devices/sensors/servers at one central place i.e. cloud will increase the vulnerability and cost. So, we need a better in-built ANN-based processing edge intelligence which will help the application or service of consumer to improve security and privacy of the data. It will also ensure less bandwidth usage, power consumption of resources, better low latency response time and intelligent allocation of network slices, etc. This leads to the neural AI-based edge computing and intelligent edge based 5th generation echo system. Analyzing noisy, ambiguous, and massive data environment like 5G needs more interpretation for high accuracy. For instance, finding out a particular match for a person by analyzing each image of different kind of thousands of cameras needs more intelligence than normal learning process.

The major types of ANN are:

Convolutional Neural Network (C-NN)
Recurrent Neural Network (R-NN)
Feed Forward Neural Network (FF-NN)
Radial Basis Function Neural Network (RBF-NN)
Long/Short term Memory Neural Network (LSTM-NN)
Modular Neural Network (M-NN)
Gated Recurrent Unit Neural Network (GRU-NN)
Hopfield Network (HN)
Boltzmann Machine Neural Network (BN-NN)

REFERENCES:

Share On

Copy link