Serial networks fail at the physical layer far more often than at the protocol layer — and the physical layer is checkable in an hour with a multimeter and honesty.
The five mistakes
- No termination, or termination everywhere. Two 120 Ω terminators, one at each end of the trunk — not on stubs, not on every device (a bus with eight terminators is a 15 Ω load no driver enjoys).
- Star topologies. RS-485 wants a daisy chain; every long stub is a reflection generator. If the layout forces a star, that is a repeater or a design change, not a hope.
- Missing biasing. An idle bus without fail-safe biasing floats; receivers read noise as start bits and you get the classic “random CRC errors when nothing transmits”. One point of bias per segment, usually at the master.
- Grounding fiction. A and B are a differential pair, but the common-mode range is finite — connect the signal common/shield per a plan (grounded at one point, drained properly), not per whatever the installer had left.
- Cable roulette. Twisted pair designed for RS-485 (correct impedance), not spare power cores, not untwisted alarm cable — and separated from VFD output cables, which are the loudest neighbors in any panel.
The one-hour audit
Power down: continuity and resistance across A–B (expect ~60 Ω with both terminators). Power up, no traffic: differential idle voltage confirms biasing. Then a scope on the far end while polling — clean edges settle arguments no meeting can.
FAQ
Maximum devices and length? Classic transceivers: 32 unit loads; modern fractions allow more. Length trades against baud rate — 9600 baud is forgiving; 115k demands the rules above be real.
Isolated converters worth it? Between buildings, across power domains, or near drives: yes, always. Isolation failures masquerade as software bugs for months.
Zone Otomasyon carries a scope to serial-bus complaints — the fix is usually visible in one screen. Bus troubleshooting.