Conventional wireless relay networks store a complete frame at each hop, then forward it. Each relay adds a full frame duration to the latency. Five hops means five frame delays, plus processing time at each node. For applications requiring sub-millisecond response — industrial control, safety systems — this accumulates past the deadline.
Note (2026-11-15; read against the paper's full text): the 0.16 percent added per extra hop is the paper's figure for 16-byte (128-bit) frames, and the essay's sentence pairs it with the 4-byte frame whose 5-hop latency is under 1 millisecond; the paper says the increment of about 5 microseconds per hop holds for frames of arbitrary size. The relays do detect each symbol (a windowed symbol detector) and relay it at once, so 'no decoding at intermediate nodes' is the essay's wording, not the paper's. The latency, reliability and scalability results come from a MATLAB simulation calibrated with small-scale software-defined-radio experiments, and the paper reports a counterintuitive result the essay omits: smaller carrier-frequency offsets cause more error bursts, and its BCH codes (80 percent code rate) hold frame loss to about 1 percent at plus or minus 1 Hz offset variation versus 0.01 percent at plus or minus 1 kHz.
Liu, Belogaev, Oostvogels, Fang, Hughes, and Famaey (arXiv:2603.23213) eliminate the store. RF-Zero-Wire relays symbols as they arrive, one by one, without waiting for the complete frame. The relay starts transmitting the first symbol of a frame while still receiving the last. No frame buffer, no decoding at intermediate nodes, just pass-through at the symbol level.
The result: a 4-byte frame traverses 5 hops in under 1 millisecond. Each additional hop adds only 0.16% to the latency, compared to over 100% per hop in store-and-forward. The latency scales with the frame length rather than with the hop count — the frame's transmission time dominates, and the relay overhead becomes invisible.
The cost is interference. Imperfect crystal oscillators at each relay create carrier frequency offsets, producing a beating pattern of constructive and destructive interference when the relayed signal overlaps with the original. The paper shows this can be mitigated with error correction coding, trading bandwidth for reliability.
The structural insight is about where the latency lives. In store-and-forward, the latency is in the relay — each hop multiplies the total. In symbol-synchronous forwarding, the latency is in the frame — each hop adds only a symbol period. Moving the bottleneck from the network to the message changes the scaling law from multiplicative to additive.