A distributed control-layer prototype exploring a contrarian premise:
Non-Terrestrial Networks (NTN) in 6G should activate under sustained sustainability stress — and not just for coverage failure, or ubiquitous coverage.
This system reframes NTN as a sustainability-aware control mechanism governed by temporal KPI degradation and deterministic policy.
Sustainability degradation in a RAN is not instantaneous. Congestion, PRB pressure, energy draw, carbon intensity, and mobility churn accumulate over time. Reactive threshold triggers are insufficient.
This architecture models sustainability as a temporal signal and governs NTN activation through a deterministic control loop.
- The model predicts risk
- Policy governs escalation
- NTN becomes a sustainability valve — not redundancy
Design principle: ML predicts sustainability stress. Deterministic policy governs NTN activation.
Inference Layer
- C++ gRPC service
- ONNX Runtime (GRU model)
- Deterministic NTN state machine with hysteresis
Control Layer
- SvelteKit API bridge
- Explicit controller arbitration
- Ordered crisis state persistence
Persistence
- Postgres-backed global system state
- Sequential temporal buffering
Infrastructure
- Docker Compose
- Dev Container workflow
- Fully reproducible environment
Strict separation of concerns:
- Prediction is probabilistic
- Policy is deterministic
- NTN activation is explainable
The interface provides:
- Controlled KPI stress simulation
- Real-time sustainability crisis trajectory
- NTN fallback state visualization
- Multi-observer / single-controller governance
The UI functions as a diagnostic and orchestration lens — not the system itself.
- congestion
- prb_util
- traffic_load
- ran_energy
- carbon_intensity
- isac_quality
- mobility_rate
Temporal input shape: (1, 60, 8)
The 8th feature is the prior crisis score, enabling closed-loop temporal feedback.
Output:
- Sustainability crisis score ∈ [0,1]
- NTN state ∈ {0,1,2,3}
NTN escalation is governed by:
- NTN_START threshold
- NTN_CROSS threshold
- Sustained critical windows
- Hysteresis-based recovery
This prevents oscillation and mirrors operator-grade control-plane logic.
NTN acts as:
- A carbon-balancing lever
- A load redistribution mechanism
- A resilience stabilizer
- A sustainability-aware orchestration layer
-
Create
.envfile:
DB_USER=<user-name>
DB_NAME=<db-name>
DB_PASSWORD=<password> -
Create
password.txtfile:
<password>
Note: this should be same as is given in the.envfile. -
Run pre-built images from docker hub:
docker compose -f docker-compose-prod.yml up -d -
Or build from source:
4.1 Build pre-requisites first:
docker compose build vcpkg-base4.2 Build and start all services:
docker compose up --build -d -
Web UI available at:
http://localhost:3000
No host toolchain required.
- Reinforcement learning for adaptive threshold tuning
- Multi-cell NTN arbitration
- Carbon-aware traffic steering policies
- GPU-accelerated inference (TensorRT)
- Federated edge inference
- Real traffic trace integration
This project sits at the intersection of:
- Distributed systems
- Temporal ML inference
- Deterministic control policy
- Multi-domain orchestration
- Sustainability-aware network design
The objective is not a model demo.
It is an executable control-layer concept for AI-native infrastructure.
This project is licensed under the MIT License - see the LICENSE file for details.
This project utilizes the following open-source components:
- PostgreSQL: Licensed under the PostgreSQL License (Permissive).
- TimescaleDB: Licensed under the Apache License 2.0 / Timescale License.
- gRPC: Licensed under the Apache License 2.0.
- ONNX Runtime: Licensed under the MIT License.
- vcpkg: Licensed under the MIT License.
- Svelte: Licensed under the MIT License.