Demonstration of a 1.2 Gbps Always-on Fully-Connected Mesh Network with RFSoC SDRs

This paper presents the first demonstration of a low-latency, always-on fully-connected mesh network using four RFSoC-based software-defined radios that achieves an aggregate throughput of 1.2 Gbps to simultaneously stream twelve real-time, uncompressed 4K video feeds over 2x2 MIMO links.

The Gist

Imagine you have four drones flying in a perfect cross shape. Now, imagine that instead of just talking to each other one by one, every single drone is talking to every other drone at the exact same time, without ever stopping to listen or waiting for a turn.

That is essentially what this paper demonstrates. The researchers built a high-tech "party" where four drones are constantly shouting their secrets to each other, and they managed to do it so efficiently that they are streaming four simultaneous, crystal-clear 4K movies (one from each drone) without any lag or pixelation.

Here is a breakdown of how they pulled off this magic trick, using some everyday analogies:

1. The Super-Brain: The RFSoC

Usually, building a radio that can do all this requires a messy pile of separate boxes: a computer for the brain, a box for the antenna, a box for the signal processing, and so on.

The researchers used a special chip called an RFSoC (Radio Frequency System-on-Chip). Think of this chip as a Swiss Army Knife that is also a supercomputer. It has the antennas built right into the brain, and it can process the radio waves instantly before they even hit the main computer. This is like having a chef who can chop, cook, and plate a meal in the time it takes you to walk to the kitchen. Because everything happens on one chip, there is almost zero delay (latency).

2. The "Always-On" Conversation

In a normal walkie-talkie, you have to push a button to talk and let go to listen. If everyone talks at once, you just hear noise.

In this experiment, the drones use a clever trick called Frequency Division. Imagine a busy highway with 12 lanes.

• Instead of everyone trying to drive in the same lane, the researchers gave each drone its own set of lanes.
• Even though they are all driving at the same time (Full Duplex), they never crash because they are in different "frequency lanes."
• The result? 12 active conversations happening simultaneously. It's like a room where 12 people are having 12 different intense conversations at once, and everyone can hear exactly who they need to hear without any background noise.

3. The "Noise-Canceling" Magic

When you shout in a crowded room, your own voice often echoes back to you, making it hard to hear others. In radio terms, this is called "self-interference."

The researchers had to build a digital noise-canceling headset inside the chip.

• The Problem: The drone's own transmitter is so loud it drowns out the signals from the other drones.
• The Solution: The chip is smart enough to know exactly what it is shouting. It subtracts its own voice from the incoming sound, leaving only the voices of the other three drones. It's like having a superpower where you can tune out your own thoughts to focus entirely on what your friends are saying.

4. The "Traffic Cop" and the "Video Stream"

To make sure the data doesn't get jumbled, the system uses a Graphical User Interface (GUI). Think of this as the Traffic Control Tower.

• It watches the "traffic" (the data) in real-time.
• If a signal gets weak (like a car hitting a pothole), the tower instantly adjusts the speed or the route to keep the video smooth.
• They measured the quality of the connection using things like "Error Vector Magnitude" (EVM), which is basically a scorecard telling them how many "typos" are in the message. They kept the typos so low (less than 1 in 100,000) that the video looked perfect.

Why Does This Matter?

The goal was to prove that we can have a resilient, high-speed network for drones that doesn't rely on cell towers or satellites.

• The Analogy: Imagine a team of firefighters, police officers, or search-and-rescue drones working in a disaster zone where cell towers are down.
• The Benefit: With this system, they can share live, high-definition video feeds from every angle instantly. They can coordinate complex movements, map the area in 3D, and make split-second decisions together, all because they are connected by this super-fast, self-healing mesh network.

In short: They turned four drones into a single, super-intelligent organism that can stream 4K video to itself in real-time, proving that the future of drone swarms is not just about flying together, but about thinking together at the speed of light.

Technical Summary

Here is a detailed technical summary of the paper "Demonstration of a 1.2 Gbps Always-on Fully-Connected Mesh Network with RFSoC SDRs."

1. Problem Statement

Networked Unmanned Aerial Vehicle (UAV) systems require resilient, high-throughput connectivity to support cooperative behaviors, synchronized operations, and real-time data exchange (e.g., surveillance, search-and-rescue). Existing solutions often struggle with:

• High Throughput & Low Latency: Supporting simultaneous, uncompressed 4K video streaming across multiple nodes.
• Full-Duplex Operation: Achieving "always-on" communication where nodes transmit and receive simultaneously without significant self-interference.
• Hardware Complexity: Traditional architectures often rely on complex analog front-ends or separate processing units for baseband and RF, increasing cost and size.
• Mobility & Interference: Maintaining link stability in dynamic environments with fading, shadowing, and adjacent-band interference.

2. Methodology

The authors designed and implemented a fully-connected (complete-graph) mesh network consisting of four UAV nodes using AMD/Xilinx Zynq UltraScale+ RFSoC ZCU111 software-defined radios (SDRs).

• Network Topology:

  • Four nodes arranged in a cross (+) configuration to provide spatial diversity and reduce shadowing.
  • Each node is equipped with a 2x2 MIMO (Multiple-Input Multiple-Output) configuration (2 transmit, 2 receive antennas).
  • This results in 12 simultaneous, always-on links (4 nodes × 3 peers each).

• Hardware Architecture:

  • Platform: ZCU111 RFSoC, integrating 8x 12-bit 4.096 GSps ADCs and 8x 14-bit 6.554 GSps DACs.
  • RF Front-End: Custom-built RF front-ends operating in the 900 MHz band with XM500 RFMC balun cards, optimized for low loss.
  • Processing:
    • Programmable Logic (PL): Handles all physical layer (PHY) tasks (modulation, MIMO combining, filtering, equalization) to ensure deterministic low latency.
    • Processing System (PS): An ARM-based core manages system control, I/O, and real-time monitoring via a host GUI.

• Signal Processing & Protocol:

  • Multiple Access: Uses Frequency-Division Multiple Access (FDMA) with adjacent frequency bands distributed to each transmitter to mitigate self-interference and enable full-duplex operation.
  • Modulation: 16-QAM over a 37.44 MHz bandwidth per link.
  • Transmitter Chain: Data streaming via AXI4-Stream DMA $\rightarrow$ Frame assembly (preamble + payload) $\rightarrow$ 16-QAM mapping $\rightarrow$ 65-tap Square-Root Raised Cosine (SRRC) pulse shaping $\rightarrow$ Digital upconversion (NCO) $\rightarrow$ DAC.
  • Receiver Chain: Direct RF sampling $\rightarrow$ Digital downconversion (NCO) $\rightarrow$ Adjacent-band filtering (3-stage low-pass filters with steep roll-off) $\rightarrow$ Matched SRRC filtering $\rightarrow$ Golay-based preamble correlation for frame detection $\rightarrow$ Adaptive symbol-spaced equalization following Maximum-Ratio Combining (MRC).
  • Equalization: Implements cascaded high-order FIR filters and adaptive equalizers to mitigate Inter-Symbol Interference (ISI) and channel selectivity caused by mobility.

3. Key Contributions

• First-of-its-Kind Demonstration: This is the first known demonstration of low-latency, digitally controlled Frequency-Division Duplex (FDD) RFSoC-based MIMO links capable of simultaneously supporting multiple real-time, uncompressed 4K video streams.
• Hardware-Software Co-Design: Successfully transitioned a Simulink-based simulation model to a hardware architecture running at a 200 MHz FPGA clock. This required extensive re-design of ADC/DAC interfaces, NCO-based frequency translation, and optimized equalizer architectures to meet strict timing constraints.
• Advanced Interference Mitigation: Implemented custom digital filtering and adaptive equalization entirely within the RFSoC's programmable logic to handle adjacent-band interference and self-interference in a full-duplex, multi-node environment.
• Real-Time Monitoring & Reconfiguration: Developed a host-side Graphical User Interface (GUI) that visualizes link performance (EVM, pre-detection SINR, BER) in real-time and allows for dynamic reconfiguration of link parameters during operation.

4. Results

• Aggregate Throughput: Achieved a total network throughput of ≈1.2 Gbps (specifically 12 links × 99.84 Mbps).
• Bandwidth Efficiency: Operated over a shared bandwidth of 200 MHz.
• Link Quality:
  • Bit Error Rate (BER): Stable operation with BER < 1e-5 across all links.
  • Signal-to-Noise Ratio (SNR): Maintained 28–30 dB across all links.
  • Error Vector Magnitude (EVM): Monitored in real-time, confirming signal integrity.
• Application Layer: Successfully demonstrated the end-to-end transmission of raw/uncompressed 4K video from each UAV, validating the system's low-latency and high-bandwidth capabilities.

5. Significance

This work represents a significant milestone in UAV communications and SDR technology. By leveraging the RFSoC architecture, the authors demonstrated that it is possible to integrate complex PHY processing (MIMO, adaptive equalization, filtering) directly into the RF sampling chain, eliminating the need for intermediate analog stages. This results in a compact, cost-effective, and highly flexible system capable of supporting the massive data rates required for next-generation autonomous UAV swarms. The ability to stream uncompressed 4K video in a fully connected, full-duplex mesh network opens new possibilities for real-time cooperative sensing, high-fidelity surveillance, and distributed AI applications in dynamic environments.