BenchLink: An SoC-Based Benchmark for Resilient Communication Links in GPS-Denied Environments

1. Problem Statement

Modern wireless communication systems rely heavily on the Global Positioning System (GPS) for accurate timing and synchronization. However, in emerging applications such as Unmanned Aerial Vehicles (UAVs), urban vehicular networks, and military operations, GPS signals are often unreliable, obstructed (e.g., urban canyons), or intentionally disrupted (spoofing/jamming).

In GPS-denied environments, the lack of a shared reference clock causes oscillator drift between transmitter and receiver nodes. This leads to:

• Carrier Frequency Offset (CFO): A mismatch in carrier frequencies causing phase rotation over time.
• Constellation Rotation: Degradation of signal quality, particularly for high-order modulations.
• Synchronization Loss: Failure to maintain link stability in dynamic, mobile scenarios.

Existing solutions often rely on General Purpose Processor (GPP)-based Software-Defined Radios (SDRs), which suffer from unpredictable latency and limited real-time processing capabilities. Furthermore, there is a lack of open-source, programmable benchmarks that allow researchers to study the trade-offs between pilot density (overhead) and modulation order (throughput) specifically under GPS-denied conditions.

2. Methodology

The authors propose BenchLink, a System-on-Chip (SoC)-based benchmark designed to operate without GPS synchronization. The methodology involves three core components:

A. Hardware/Software Co-Design Architecture

• Platform: Implemented on Zynq UltraScale+ MPSoC (ZCU102) boards, integrating ARM Processing Systems (PS) and Field-Programmable Gate Arrays (FPGA).
• Partitioning:
  • ARM PS: Manages high-level protocols, RF control parameters, and data interfacing with an onboard Intel NUC.
  • FPGA (Programmable Logic): Handles computationally intensive Physical Layer (PHY) tasks (baseband processing, pulse shaping, CFO correction) to ensure deterministic, low-latency performance.
• RF Front-End: Uses AD-FMCOMMS3-EBZ boards with AD9361 transceivers operating at 2.485 GHz.

B. Adaptive Frame Design & CFO Mitigation

To combat clock drift without GPS, BenchLink employs a two-stage synchronization strategy:

1. Coarse CFO Correction: Uses a fixed preamble with autocorrelation properties (Golay Complementary Sequences) to estimate and correct the initial frequency offset at the start of every frame.
2. Adaptive Fine-Tuning: Introduces a configurable pilot density mechanism.
   • The frame structure allows the number of pilot sequences per packet ($\lambda_p$) to be dynamically adjusted.
   • Trade-off Logic: Higher $\lambda_p$ improves channel estimation and residual CFO correction (crucial for high-order modulations and fast-fading channels) but reduces data throughput due to overhead. Lower $\lambda_p$ increases throughput but risks link failure in dynamic environments.
   • The system uses an asynchronous FIFO and a state machine to assemble frames dynamically based on the configured $\lambda_p$.

C. Experimental Testbeds

The system was evaluated in three distinct scenarios:

1. Indoor (Controlled): Wired connections to isolate oscillator drift from channel effects.
2. Ground-to-Ground (G2G): Portable carts with BenchLink modules in an outdoor environment.
3. Air-to-Air (A2A): Heavy-lift drones carrying BenchLink modules, hovering at 12m altitude with 30m separation.

3. Key Contributions

1. SoC-Based Benchmark Design: A fully open-source, reconfigurable wireless link design that leverages FPGA acceleration for low-latency, deterministic PHY processing, overcoming the limitations of GPP-based SDRs.
2. Adaptive Pilot Mechanism: A novel frame structure that allows real-time reconfiguration of pilot density ($\lambda_p$) to adapt to varying channel coherence times and oscillator drift rates.
3. Comprehensive Dataset: Collection and release of a large-scale dataset (18 MB, ~108,000 rows) covering indoor, G2G, and A2A scenarios. The data includes residual frequency offset, goodput, EVM, SINR, and metadata conforming to the Signal Metadata Format (SigMF).
4. Empirical Validation: Extensive experimental evaluation demonstrating the system's resilience in GPS-denied environments across various modulation schemes (4QAM to 64QAM).

4. Key Results

The experiments revealed critical insights into the relationship between pilot overhead, modulation, and environment:

• Impact of Pilot Repetition ($\lambda_p$):
  • Indoor: Increasing pilots from 1 to 8 reduced residual phase offset from ~9.8° to <1.8°, with diminishing returns after 4 repetitions.
  • G2G vs. A2A: A2A links (UAVs) suffer from shorter channel coherence times due to Doppler effects and jitter, requiring more frequent pilot insertion than G2G links.
• Goodput Performance:
  • Low-Order Modulation (4QAM/8QAM): Increasing pilots often resulted in reduced goodput because the overhead outweighed the marginal gain in channel estimation accuracy.
  • High-Order Modulation (16QAM/64QAM): Significant goodput gains were observed with increased pilots.
    • G2G Example: 16QAM goodput increased by 356% (102 kbps to 468 kbps) when moving from 1 to 4 pilots.
    • A2A Example: 64QAM goodput increased by 1200% (33 kbps to 424 kbps) when moving from 1 to 6 pilots.
• Signal Quality:
  • EVM (Error Vector Magnitude): Generally decreased (improved) with more pilots, though a slight dip in performance was observed at $\lambda_p=2$ for lower-order modulations in G2G due to multipath fading.
  • SINR: Improved monotonically with pilot density, particularly in A2A scenarios where line-of-sight is dominant.

5. Significance and Future Work

Significance:

• Enabling Resilient Connectivity: BenchLink provides a critical tool for developing communication systems that function reliably in GPS-denied or contested environments, which is vital for military, disaster response, and urban air mobility.
• Research Catalyst: By releasing the hardware design and a rich, labeled dataset, the authors enable the wireless community to develop and test data-driven link adaptation algorithms (e.g., using Machine Learning) to autonomously optimize pilot density and modulation in real-time.
• ISAC Support: The low-latency SoC architecture makes the platform suitable for Integrated Sensing and Communications (ISAC) applications where precise timing is required.

Future Work:

• Learning-Based Optimization: Using the collected dataset to train ML models for autonomous, real-time link configuration.
• Protocol Enhancements: Integrating Forward Error Correction (FEC) and Automatic Repeat Request (ARQ) to improve goodput further.
• Scalability: Expanding testing to large-scale multi-UAV networks and interference-rich environments (e.g., jamming scenarios).