5G O-RAN (Open RAN): Architecture, Procedures And Use Cases

Understand the traditional base station architecture for 2g–4g, where the core network connects to a cabinet containing baseband and radio units, with baseband processing before modulation and RF cable losses.

The lecture traces 4G LTE's shift from traditional to contemporary base stations, moving the radio unit near the antenna as a remote radio unit and linking via optical fibre fronthaul.

Explore how evolution to virtualized ran decouples hardware and software, enabling virtualized bpu functions to run on shared servers with cots hardware via a virtualization layer, while fronthaul remains proprietary.

Explore how vRAN evolves toward open RAN by using software defined radio on commercial off-the-shelf hardware. Replace proprietary R2 hardware and the proprietary interface with an open interface.

Compare distributed ran and centralized ran, noting that processing near the antenna reduces front hall costs, while centralizing BPU enables resource pooling for stadium and residential users.

Compare decentralised RAN and cloud RAN. Baseband units sit near the core, while cloud RAN deploys software-based virtual network functions on the cloud, scalable and capable of resource sharing.

Explore how 3GPP 38.801 splits the baseband unit into decentralised and distributed units to enable horizontal dis-aggregation, optimizing placement near core or remote units.

Examine how ran functionalities split across the central unit, distributed unit, and remote radio unit, and compare pdcp, rlc, mac, phy, and rf layer splits and their pros and cons.

RAN split option 1 centralizes processing within the radio access network, lowers front-part data rate, and allows higher latency, with microwave links possible but increased power, size, and cost.

RAN split option 2 centralizes the BTC B layer while most functions stay in the remote radio unit, enabling low latency and dual connectivity via a centralized DCP layer.

Explain how 5g o-ran split option 8 centralizes four layers, keeps the rf layer in the remote unit, and enables virtualization with header additions, arq timing, and five-millisecond roundtrip constraints.

RAN split option 6, adopted by the Small Cell Forum, centralizes DP, DCP, oral, C and D Mac layers at a centralized location, with physical and RF layers implemented accordingly.

Explore RAN splits and front hall tradeoffs: higher level splits favor rural coverage; lower level splits favor dense urban capacity with high data rates and low latency.

Explore the 7.2x split option for Open RAN, with recording and beamforming in the low physical layer and virtualization of MAC, RLC, DCP, and outer layers.

Explore the enhanced CPRI (e-CPRI) and ECB fronthaul protocols, a packet-based open interface that uses Ethernet LANs to reduce optical fiber needs and improve fronthaul efficiency in urban indoor deployments.

The lecture explains the difference between open ran and Orient, noting open ran as disaggregated hardware and software with open interfaces, and Orient as the Orient Alliance's architecture and specifications.

Disaggregate the control plane and user plane to independently scale resources, move the rrc to the control plane and pdcp to the user plane, enabling edge computing and ultra-low latency.

Explore the O-RAN architecture by detailing service management and orchestration, the RAN intelligent controller with non-real-time and near real-time components, E2 interfaces, and AR and X apps.

Explore aggregation options for oran nodes by combining cu control plane, du reserve plane, and distributed unit into one entity with various interfaces, and note the trade-offs of shared interfaces.

Explore how O-RAN control loops operate at real-time, near real-time, and non-real-time levels, executing within ten milliseconds, providing action feedback to beamforming and scheduling decisions.

Identify major players shaping the O-RAN ecosystem, including 3GPP baseline specs, the O-RAN Alliance, Linux Foundation software community, Open Networking Foundation's SDN/Norse Rig project, and the Telecom Infra Project.

Trace the evolution from physical network functions to virtual network functions and cloud network functions, comparing architectures and highlighting PNFs’ bulky hardware, high costs, and limited scalability.

Leverage virtualization with virtual network functions on commercial off-the-shelf hardware; a hypervisor creates virtual machines with separate operating systems for software-based bbu functions, though boot is slow and storage intensive.

In the BNF approach, monolithic applications feature tightly coupled modules, creating interdependencies where a single module failure or update affects the whole app, deployed on a single virtual server.

Adopt cloud native network functions by deploying distributed microservices, loosely coupled with their own databases, running in lightweight containers and supporting multiple instances for uninterrupted operation.

Analyze regional and edge cloud deployment options for 5G O-RAN, considering data center locations, telco rooms, and capacity limits to place near real-time RAN and SDR hardware.

Explore how devops enables continuous integration and deployment of cloud-native microservices in O-RAN containers, allowing independent testing, rapid updates, and a feedback loop between development and operations.

Explore how network automation reduces capex and opex by automating cloud setup, zero-touch provisioning, and automated testing and upgrades, with ai and ml optimizing performance.

Explore how the service management and orchestration manages cloud-based entities via an open interface, covering fault detection and correction, configuration, accounting, collection of key performance indicators, and analytics.

Initiate policy-based guidance and an EIA AML model from the non real time RIC to the near real time RIC via the event interface to optimize and control RAN functions.

Explore the JSON format through a nested example, detailing name-value pairs, and objects like address and profile with fields such as street, city, state, postal code, designation, and department.

Understand event policies from non-real time to near real time, including policy IDs, scopes, and policy statements. Apply examples to set QoS parameters for users, slices, and traffic types.

Explore near real-time RIC for admission control, mobility management, and interference management to optimize radio resources in 5G O-RAN architecture.

Explore centralized and distributed near-real-time RIC deployments that connect to multiple E2 nodes across cells, collect KPI data, and optimize cell performance in a coordinated way.

Explore how the O-RAN traffic steering app enables user equipment centric policies for fine-grained control of UE services, improving on traditional TS where all UEs were treated the same.

Explore traffic steering with open radio access networks that enable user equipment–centric policies for tailored handover and service priorities, enhanced by machine learning for network and device performance.

Identify data required for traffic steering, including measurement reports and signal quality from user equipment, plus handover and cell load KPIs, with near real-time KPIs and capas data at frequencies.

Non real time RIC defines traffic steering policy and sends enrichment data to the near real time RIC, using analytics to generate the ATF fingerprint and RF quality measurements.

The near real-time RIC receives policy from the non real-time RIC and implements traffic steering, using coverage and quality maps to send E2 messages for handovers.

Track how two E2 nodes receive and execute control messages, collect real-time KPIs, and transmit them via the E2 interface to the near real-time rig, enabling action assessment and modification.

This example scenario explains traffic steering between two GNB base stations with different bands and latencies, routing voice QoS 1 and data QoS 9 via near real-time policies.

Explore how a single physical 5G network hosts multiple network slices, each with distinct requirements for mobile broadband, massive IoT, and mission critical IoT, isolated yet sharing resources.

This 5G network slicing example shows a user equipment using both the voice over 5G slice and the enhanced mobile broadband slice to access data networks, with virtualized network functions.

Describe how S-NSSAI identifies a network slice with an 8-bit slice type and a 24-bit differentiator, mapping 1–4 to enhanced mobile broadband, reliable low latency, massive IoT, and operator-defined types.

Explain how a network slice is built from multiple network slice subnet instances, each containing network functions and possibly shared across slices, including core and access network functions.

Describe how customer demands flow through the communication service provider to the network slice provider and subnet provider, using CSMF, NSMF, and NSSMF to translate network type, capacity, and QoS.

Understand the network slice template, which lists resources and attributes for a network slice; reuse or scale an existing template to meet customer requirements.

ONAP architecture enables network slicing through design-time and runtime environments, where design-time provides functions and a visual tool to model assets and runtime enforces policies via an inventory and orchestrator.

An app-based o-ran slicing architecture centers on service orchestrator coordinating communication service management, network slice management, and subnet management, with owner optimization framework providing network slice templates and subnet instances.

Learn to create a network slice using ONAP-based architecture. Users submit requirements via the CSM portal, which converts them to slice specs and provisions them via the optimization framework.

This use case applies AI-driven traffic demand prediction to optimize resource allocation for network slices in O-RAN, considering time, location, and application patterns, with data from in-nodes.

Collect KPI data via the Auvergne interface, train an ML model to predict subnet resource needs, and optimize network slice subnet instances by reconfiguring nodes and updating cloud resources.

Collect performance data from two NSA-site nodes during peak times, train an AML model, and use predictions to allocate ITU and cloud resources via the depicted interfaces.

Dynamically allocate UAV radio resources along its flight path to hand over coverage between base stations via beamforming, while managing side lobes and uplink interference.

Use server and SMU data to predict UAV uplink/downlink interference and allocate optimal cell, beam, bandwidth, and numerology, updating AML model and radio access network resources via near real-time interfaces.

Depict a flow diagram of flight-path based dynamic UAV radio resource allocation in 5G O-RAN, using near real-time and enrichment data to optimize UAV coverage.

Examine 5G massive MIMO beamforming optimization, including codebook and non-codebook beams, grid sweeping, and how O-RAN enables centralized power and resource optimization across multi-vendor cells.

Explore how cell site information, intercept distance, operating frequency, bandwidth, and beam configuration feed a nonlinear model to predict massive MIMO beam setups for non-realtime optimization.

Explore how ai and ml can improve network performance and outline a workflow for implementing ai and ml, differentiating the two and highlighting learning from data.

Learn how machine learning enables computers to learn from experience without explicit programming, focusing on supervised learning, its classification and prediction, and the importance of labeled data.

Explore linear regression and its application by modeling output throughput as a function of base station transmit power, using beta naught and beta one to train a predictive model.

Explore neural networks, including neurons and layered architecture, for image classification. Learn how input pixels map to hidden and output layers to label images as peach, apple, or beer.

Explore how back propagation uses gradient descent to train neural networks, adjusting weight coefficients to minimize the classification error and reach the minimum error.

Show how neural networks in mobile networks classify load as low, medium, or high using inputs from four base stations and time, trained on labeled data to predict future load.

Apply logistic regression for binary classification by mapping reference signal received power to coverage probability with a sigmoid function, train with the gradient descent algorithm, and use a 0.5 threshold.

Explore unsupervised learning with unlabeled data, using k-means clustering to group data points into clusters based on patterns; the number of clusters is chosen by the user.

Learn the four stages of the K-means clustering algorithm: initialization, assignment, update, and iterative refinement, where randomly placed centroids are reevaluated until all data points belong to the nearest centroid.

This example shows how k-means clustering classifies reference signal received power values into three clusters, revealing low, medium, and good coverage areas in a base station's mobile network.

Learn reinforcement learning, where a learning agent interacts with an environment, takes actions, receives rewards (positive, negative, or zero), and discovers an optimal policy to maximize reward.

Explore reinforcement learning in three base stations learning transmit power policies to maximize throughput and learning traffic steering via handover margins to reduce call blocking.

Explore how AI and ML analyze vast radio access network data and alarms to diagnose issues, predict security threats, optimize resource utilization, save energy, and forecast hardware failures.

Explore the ai/ml framework for O-RAN, detailing data sources from OR, ODU/OCU, core network, and applications, and data collection via open fronthaul plane interfaces and O-1/E-2 interfaces to train algorithms.

The AI/ML framework for O-RAN routes data from collection to the training host, model management, and inference host, enabling offline and online training, certificate attachment, catalog deployment, and parallel inference.

The actor runs ML model from inference host to drive ML-assisted decisions over O one and E two interfaces, using data from OCU, ODU, and ORU to refine the model.

Map ai-ml functionalities into O-RAN control loops by placing training and inference blocks in loop one, loop two, or loop three based on latency and availability from o1 and e2.

Define deployment scenario 1.1: data from the O1 interface trains an AIML model, with O1 or A1 actions, then training, certificate, cataloging, and inference with continuous performance feedback.

Deployment scenario 1.2 trains and certifies models in SMO, then runs near real-time inference on E two interface using inputs from O one to predict outcomes with low latency.

Compare deployment scenario 1.3 with 1.1 to show blocks can be inside the non-real-time rake or outside it, illustrating flexible placement in 5G o-RAN architecture.

Deployment scenario 1.4 uses offline SMO training in non real time and online near real time training, with ML inference on OCU/ODU data via the O1 and E2 interfaces.

Illustrates the end-to-end machine learning model life cycle, from initial design with tools like scikit-learn, Keras, TensorFlow, to training on a data lake and dockerized deployment.