Testbed Evaluation of Distributed Radio Timing Alignment Over Ethernet Fronthaul Networks

Abstract

This work investigates a mechanism for alignment of the timing on which spatially distributed and cooperative radio units transmit in radio-frequency (RF) when served over a packet-based fronthaul. It analyzes the problem by considering the imperfect clock synchronization of the radio units and the packet delay variation that fronthaul packets are subject to. Following the analysis, this paper proposes an implementation architecture for distributed RF transmission timing alignment based on synchronized triggering among radio units and centralized processing units. Throughout this discussion, special attention is given to the scheme's impact on the overall achievable fronthaul latency. Subsequently, this work discusses both hardware and software aspects of a prototype that was developed based on field-programmable gate arrays (FPGAs). In the end, it presents results obtained on an Ethernet fronthaul testbed where the referred FPGA-based prototypes implement radio units that are synchronized using the IEEE 1588 precision time protocol or by pulse-per-second references. Results validate the functionality of the proposed architecture and illustrate various relevant choices concerning system parameters. INDEX TERMS 5G, clock synchronization, Ethernet, fronthaul, precision time protocol.

Clock Synchronization Algorithms Over PTP-Unaware Networks: Reproducible Comparison Using an FPGA Testbed

“…In this experiment, the s-to-m delays (bottom right corner in Fig. 12) experience peak collisions close to the worst-case delays predicted by (53). Correspondingly, the static delay asymmetries (summarized in Table 8) are relatively high, given that they include the average queuing delays.…”

“…Fig 4 shows the FPGAs, the switches, and several measurement instruments that compose the testbed, some of which are out of scope. The interested reader can refer to the description of an early version of this testbed in [52] and a more recent usage regarding radio-frequency timing alignment experiments in [53].…”

“…can experience an extra queuing of τ bg . Thus, under the ZG condition, the worst-case s-to-m delay becomes: (53) where β sm [j] models the static hardware latency associated with the j-th hop in the s-to-m direction.…”

“…Equations (52) and (53) define the worst-case (highest) expected PTP delays. However, the PTP messages can experience significantly lower delays when departing with the maximum possible BG-PTP gap and in the absence of contention.…”

“…In this scenario, we can anticipate the worst-case PTP delays using (52) and (53). For simplicity, we can assume a constant processing delay per hop of γ[j, n] = 3.3 µs and a constant hardware latency per hop of β ms [j] = β sm [j] = 300 ns (see, e.g., [23]).…”