Effective throughput of AX.25 protocol

Bartosz ZielińskiBulletin 2013 · DOI

Abstract

AX.25 protocol belongs to the HDLC protocols family and is used, among others, as a data link layer in an amateur Packet Radio network. The paper presents a simple analytical model developed on the basis of protocol frame exchange rules. Then, we compare analytical results to the experimental ones achieved using various different types of TNC controllers that act as network interfaces for Packet Radio network. In the experimental part we focus on both hardware and software properties of TNC that have influence on the effective throughput of the network.

Introduction

The AX.25 protocol has served as the backbone of amateur packet radio and remote telemetry since the 1980s. As a derivative of the High-Level Data Link Control (HDLC) family, it provides a robust framework for data transmission over radio links. However, a persistent gap exists between the theoretical performance predicted by mathematical models and the actual throughput observed in real-world deployments. This paper bridges that gap by investigating the hardware and firmware constraints of Terminal Node Controllers (TNCs)—the microprocessors responsible for implementing the protocol.

Analytical Modeling of the AX.25 Cycle

A key contribution of this work is a refined analytical model for the AX.25 transmission cycle. Traditional models often assume "perfect" hardware with negligible processing time. In contrast, this study incorporates realistic parameters such as transmitter key-up delay ($T_{CS}$) and synchronization time ($T_{103}$). Furthermore, the model applies a bit-stuffing transparency factor of 63/62 to the entire frame, providing a more accurate estimation of overhead than models that only consider the data field.

Figure 4: Timing diagrams for AX.25 frame exchange showing single frame vs. burst transmission.

Hardware Bottlenecks and the "Invisible" Window

The research highlights how hardware constraints create an "invisible" ceiling on window sizes. While a user might configure a window size ($k$) of 7, the TNC's CPU speed and internal buffering often limit the actual number of frames sent in a single burst. For instance, older Z80-based controllers may only manage to send 3 or 4 frames before the protocol timers or buffer limits force an acknowledgment cycle. Firmware logic also plays a critical role; some implementations wait for a specific timer ($T_2$) to expire before acknowledging, which significantly degrades throughput in half-duplex environments.

Figure 5: Pipelined frame transmission and acknowledgment timing.

Experimental Results and Performance Benchmarks

To validate the model, various TNCs (utilizing Z80, 68HC11, and 68302 CPUs) were tested over a wired "perfect channel" to eliminate radio interference. The results were striking:

  • 9.6 kbps Links: High-end controllers like the DLC7 reached near-theoretical throughput (~5.5 kbps), while legacy Z80 models struggled to hit 1.5 kbps.
  • 38.4 kbps Links: The performance gap widened significantly. Modern controllers (TNC3, TNC7) achieved 8–11 kbps, while legacy hardware remained capped at 5 kbps, proving that the CPU was the primary limiting factor.
  • Firmware Impact: "The Firmware" (TF) version 2.7 was found to be the most efficient for Z80 chips, but only when the clock speed was high enough (e.g., 10-20 MHz).

Conclusion

The study concludes that for high-speed amateur radio links, the processing power and firmware efficiency of the TNC are more critical than the protocol parameters themselves. Future work will focus on Software Defined Radio (SDR) implementations, where high-speed PC processors can eliminate the hardware bottlenecks identified in this study, and on optimizing protocols for low-latency real-time applications.

Summary updated

Abstract

Abstract AX.25 protocol belongs to the HDLC protocols family and is used, among others, as a data link layer in an amateur Packet Radio network. The paper presents a simple analytical model developed on the basis of protocol frame exchange rules. Then, we compare analytical results to the experimental ones achieved using various different types of TNC controllers that act as network interfaces for Packet Radio network. In the experimental part we focus on both hardware and software properties of TNC that have influence on the effective throughput of the network.
BibTeX
@article{zielinski2013effective,
  author = {Bartosz Zieliński},
  title = {{Effective throughput of AX.25 protocol}},
  journal = {Bulletin},
  year = {2013},
  doi = {10.2478/bpasts-2013-0068},
}
Figure