Documentation

Welcome to the TOSSI documentation. TOSSI is the Telecom Open Source System Integrator, a neutral foundation that integrates the open telecom stack into a coherent, validated whole. These docs cover the integration work itself: making components from different projects run as one system, and proving it with testbeds and certification.

New here?

The xFAPI bridge is the clearest example of what TOSSI does: an L1 and an L2 from different projects, made to interoperate across a standard interface. For the certification side, start with Cloud Native Telecom Certification.

The FAPI split

A disaggregated FAPI interface with a shared-memory transport, so an L1 from one project can drive an L2 from another across the standard L1/L2 boundary.

OCUDU L1 with OCUDU L2 The xFAPI bridge itself: how the interface and its transport work.

OAI L1 with OCUDU L2 End-to-end interoperability across the FAPI boundary between two open stacks.

NVIDIA Aerial L1 with OCUDU L2 A commercial GPU L1 driving an open L2 over the same interface.

NVIDIA Aerial L1 with OAI L2 The same Aerial L1, paired instead with the OAI L2.

Installing NVIDIA Aerial L1 Getting Aerial L1 built and running before you connect an L2 to it.

Hardware acceleration

Moving the expensive parts of the physical layer onto GPUs and accelerator cards, without taking the data path out of line.

Inline GPU: PRACH and SRS Inline GPU processing in the AI-RAN data path, without leaving the pipeline.

PRACH GPU Offload Offloading PRACH detection to the GPU using CUDA graphs.

Intel ACC100 LDPC Offload Moving LDPC encode and decode for PDSCH and PUSCH onto an accelerator card.

AI-RAN

Machine learning placed inside the scheduler and the link adaptation loop, trained offline and evaluated per UE at runtime.

ML-Based UL MCS Learned uplink modulation and coding selection, adapting per UE instead of by fixed table.

ML-Based BSR Periodicity Predicting each UE's next uplink arrival to tune its buffer status report timer.

Predictive CSI for Link Adaptation Predicting channel state ahead of time so link adaptation stops reacting late.

Cell Congestion Optimization rApp A closed control loop from a live RAN through the SMO and Non-RT RIC and back.

Digital twins with Sionna RT Deterministic ray-traced RF simulation driving a real RAN stack.

Core and datapath

The user plane moved into the Linux kernel, so forwarding, metering and policy enforcement all happen in the fast path.

SD-Core with eUPF An eBPF/XDP user plane with PFCP-driven subscriber metering and in-kernel quota enforcement.

Magma eBPF Deployment Guide Deploying Magma with an eBPF GTP-U datapath in place of the kernel GTP module.

Non-Terrestrial Networks

3GPP Release 17 NTN, with the satellite link modelled in software so the timing can be exercised without a satellite or any RF hardware.

NTN: An Introduction What Release 17 changes for timing, Doppler and the RACH procedure.

GEO NTN Testbed A full 5G Standalone network over a modelled 240 ms geostationary link.

LEO NTN Testbed The low-earth-orbit case, with a moving satellite and continuously changing delay.

UE and testing

Multi-UE Simulation Scale from a single UE to many over ZMQ, with no radio hardware.

End-to-End Network Slicing A slice carried the whole way from the UE through the RAN to the core.

Inter-gNB Xn Handover Direct RAN-to-RAN handover between two gNBs, with KPIs exported over O1.

Certification

CNTC turns published 3GPP and SCAS requirements into automated tests, then into a per-function pass or fail certificate, run against your own live deployment.

Cloud Native Telecom Certification The certification model: how requirements become tests, and tests become a verdict.

UPF Certification Certify a User Plane Function as a black box over its N3 and N4 interfaces.

Control-Plane Certification 44 automated tests across AMF, SMF, NRF, AUSF and UDM, each anchored to a 3GPP protocol and SCAS spec.

How these docs are organized

We follow the Diátaxis model. Within each area you will typically find an overview that orients you, an architecture page explaining how the layer works, a how-to with concrete procedures, and a reference for interfaces and standards.

Everything documented here is integrated and validated in the open. Where a result depends on a specific rig or deployment, the page says so.