Objective
Why Communication Choices Matter on a Plant Floor
An office network lives in a quiet, air-conditioned building. Yours doesn't. The cable tray you're pulling comm cable through runs past variable frequency drives, motor starters, contactors slamming in and out, welding machines, and a few hundred feet of three-phase power. All of that is electrical noise, and all of it is trying to get into your signal.
What's Actually Working Against You
- VFD output leads. A drive chops DC into pulses with very fast rise times. Those fast edges radiate, and they couple capacitively and inductively into anything running alongside them. VFD output cable is the worst neighbor a comm cable can have.
- Contactors and motor starters. Every time a coil drops out, the collapsing magnetic field kicks a voltage spike back into the wiring.
- Long cable runs. The longer the wire, the better an antenna it is, and the more resistance and capacitance sit between the driver and the receiver.
- Ground potential differences. The "ground" at the MCC and the "ground" out at a remote skid are not necessarily the same voltage. Across a large plant they can differ by volts, and that difference shows up on any signal referenced to ground.
Where We're Going
RS-232 first (simple, and easy to break), then the single-ended vs. differential noise story that is the heart of this lesson, then RS-485 in practice, ground loops, topologies, the difference between a physical layer and a protocol, and finally Ethernet and DeviceNet, a comparison table, and what actually breaks these things in the field.
Serial Standard
RS-232 — Single-Ended and Ground-Referenced
RS-232 (formally TIA/EIA-232) is the oldest of these and the simplest to understand. One driver, one receiver, one signal wire per direction, and a shared ground between the two ends. The receiver decides whether a bit is a one or a zero by measuring the voltage on the signal conductor with respect to that shared ground. That single fact is the source of every strength and every weakness the standard has.
The Numbers
- Signaling: single-ended — one conductor per signal, referenced to ground.
- Voltage levels: the standard allows driver output anywhere from ±3 V to ±15 V; real transceivers typically put out roughly ±5 V to ±12 V. A receiver treats anything more negative than −3 V as a logic 1 (mark) and anything more positive than +3 V as a logic 0 (space). Note that RS-232 is inverted — negative is the one.
- The dead band: between −3 V and +3 V the level is undefined. A receiver is not required to make sense of anything in that window.
- Topology: point-to-point only. One driver, one receiver, per signal. No addressing, no multidrop.
- Distance: commonly quoted at about 50 ft (15 m). Treat that as a rule of thumb, not a hard spec — the standard actually limits total load capacitance (2500 pF), so low-capacitance cable at a low baud rate can go considerably further, and cheap cable at high baud may not even make 50 ft.
- Speed: traditionally up to 19.2 kbps, though modern transceivers commonly run much faster over short cable.
The Voltage Picture — Valid One, Valid Zero, and the Dead Band Between
The two valid regions and the undefined window between them. Nothing sitting in the dead band is a legal one or a legal zero.
Three Things to Take From That Picture
- RS-232 is inverted, and it surprises everybody. A negative voltage is a logic 1 (mark) and a positive voltage is a logic 0 (space). Put a meter on an idle RS-232 transmit line and you should read a steady negative voltage — idle is marking. That's normal, not a fault.
- The dead band is deliberate noise margin. The driver has to put out at least ±5 V while the receiver only needs ±3 V to decide, so a couple of volts of noise can ride on the line and the bit is still read correctly. The undefined window is what buys that margin — the receiver simply refuses to guess in the middle rather than flipping back and forth on noise.
- That same dead band is why long runs get flaky. Every foot of cable adds capacitance. The driver must charge that capacitance through its own output impedance, so the edges stop being vertical and start sloping — they slew. A slow edge spends real time crossing the −3 V to +3 V window, and while it's in there the receiver has no defined answer. Add a little noise on top and bits get sampled wrong. It's why the same cable works at 9600 baud and falls apart at 115.2 k, and why the standard's real limit is 2500 pF of load capacitance rather than a number of feet.
The Concept That Matters Most
Single-Ended vs. Differential — How Noise Gets Cancelled
This is the page to remember. RS-232 and RS-485 carry the same ones and zeros. The difference is what the receiver measures, and that difference is the entire reason one of them dies in a noisy plant and the other one doesn't.
First — What Does a Noise Spike Actually Look Like?
Before we cancel it, look at the thing itself. Noise is just unwanted voltage, added to your signal, at the moment it couples in. On a scope it shows up as a fast, short-lived spike riding on top of an otherwise clean waveform.
A noise spike is unwanted voltage added to your signal. If it's big enough to push the waveform across the receiver's decision threshold, the receiver reads a bit that was never sent.
Now — Two Ways to Read That Signal
Say It in Words
Serial Standard
RS-485 in Practice (and a Word on RS-422)
You now know why RS-485 is tough. Here's what you actually have to know to install and troubleshoot one.
The Numbers
- Signaling: differential, one twisted pair per direction. The receiver detects a difference as small as ±200 mV between A and B — the driver puts out far more than that, and that gap is your noise margin.
- Common-mode range: roughly −7 V to +12 V. Noise or ground offset that shifts both conductors together inside that window is rejected outright.
- Distance: up to about 4000 ft (1200 m). That's the classic figure and it assumes a low data rate. Distance and speed trade against each other: 10 Mbps over a short run, or roughly 100 kbps out at 1200 m — not both.
- Nodes: the original standard supports 32 unit loads on a segment. "Unit load" is an electrical loading figure, not literally a device count — many modern transceivers are rated 1/2, 1/4, or 1/8 unit load, so you'll see 64, 128, or 256 devices on a segment with those parts. Repeaters extend it further.
- Two-wire (half duplex): one pair shared by everybody. Only one device transmits at a time and the protocol decides whose turn it is — which is why Modbus RTU on RS-485 is master/slave with addressing.
- Four-wire (full duplex): two pairs, one per direction. The master transmits on one pair and listens on the other, so devices can talk and listen at once.
- Termination: a 120 Ω resistor at each end of the trunk — two total, no more, no fewer. It matches the characteristic impedance of the twisted pair so the signal is absorbed at the ends instead of reflecting back down the line and smearing the data.
- Biasing: when no device is driving the line, the pair floats and receivers can chatter on noise. Bias (pull-up/pull-down) resistors hold A and B apart by a defined idle voltage so the line reads as a clean idle state. Many devices have this built in and jumper-selectable.
RS-422 — the Close Cousin
RS-422 is also differential, also twisted pair, and also good for about 4000 ft, so it cancels noise exactly the same way. The difference is topology: RS-422 is one driver talking to up to 10 receivers — broadcast, not multidrop. RS-485 added the ability for multiple drivers to share the line by turning their transmitters off when it isn't their turn. Remember it as "422 = one talker, 485 = many talkers."
All of these name the same two terminals, and vendors are famously inconsistent about which one they call "A." If a link won't talk, swapping the pair is a legitimate two-minute test — see the troubleshooting page.
Grounding
Ground Loops — The Quiet Killer on Any Comm Circuit
Ground is not one thing. It's a lot of separate rods, structural steel, conduit, and green wires that we assume are all at the same potential. Over any real distance, they aren't. Two panels grounded at different points in a plant can sit at slightly different potentials — usually millivolts to a few volts, and considerably more during a fault or a big motor start.
Why It Hurts — and Why Differential Helps So Much
On a single-ended link (RS-232), the receiver's reference is ground. So a ground potential difference between the two ends isn't merely noise near the signal — it is signal error, one for one. A 2 V offset between panels is a 2 V error added to a signal whose decision threshold is ±3 V. There is no mechanism to reject it.
On a differential link, the ground offset shifts both conductors together. The receiver reads only A − B, so a shift applied equally to A and B cancels in exactly the same way a noise spike does. That is the whole reason RS-485 works between buildings and RS-232 doesn't.
Ethernet Goes One Step Further: Transformer Isolation
10/100/1000BASE-T isn't only differential — every port also has magnetics (a small isolation transformer) between the connector and the chip's transceiver. The signal crosses that transformer magnetically, so there is no DC path from one panel's electronics to the other's. That's galvanic isolation, and it's the reason Ethernet is notably tolerant of ground potential differences between panels: with no continuous copper path for the loop current, the loop largely doesn't form in the first place.
Those magnetics are rated for a limited isolation voltage, though. When two ends are far apart, on separate ground systems, or in a high-fault-current area, the correct answer is fiber — glass carries no current at all, so the two ends are completely decoupled electrically. That's why plant standards so often say "fiber between buildings."
Mitigations Technicians Actually Use
- Terminate the shield at ONE end only — this is called single-point or single-ended grounding. Normally the control panel end. Bonding both ends is what closes the loop through your cable. Carry the shield through, isolated, at every junction box. The next page covers shield types and shield termination in detail.
- Use differential signaling wherever the run leaves a cabinet — RS-485 or Ethernet instead of RS-232.
- Galvanic isolation: opto-isolated RS-232/RS-485 converters, isolated repeaters, and isolated I/O modules break the DC path deliberately. These are cheap and are the standard fix for a stubborn serial link.
- Fiber when the two ends are far apart or on separate ground systems. Complete isolation, plus it solves distance at the same time.
- Equipotential bonding: tie the ground systems together properly with a low-impedance bonding conductor so the offset never develops in the first place. This is the real fix, but it's a design/electrical job, not something you patch at the panel.
- Keep comm cable out of trays with VFD output leads. Separate tray or conduit; cross power at 90°, never run parallel.
You've seen this idea before: 4–20 mA. Analog instrument loops dodge a lot of the same trouble by carrying information as current instead of voltage. Current is the same everywhere in a series loop, so voltage drop along a long run doesn't change the reading, and a ground offset that would wreck a 1–5 V signal doesn't change the current. It also gives you live-zero diagnostics — 0 mA means a broken loop, not a valid 0%.
It's a useful tie-back because the logic is familiar: pick a signaling scheme whose quantity isn't the one the noise is corrupting. But 4–20 mA is an analog instrument signal, not a communication network — it carries one value, not messages, addresses, or multiple devices. For this lesson it's context, not content.
Noise Immunity
Shielding — Types, Termination, and Single-Point Grounding
Differential signaling cancels the noise that gets onto the pair. Shielding keeps a large share of it from getting on there in the first place, and the twist handles a part that the shield can't touch at all. Those are three separate mechanisms, and a technician who knows which one does what stops guessing on a noisy link.
Two Ways Noise Couples In — and They Have Different Cures
- Capacitive (electric-field) coupling. A noisy conductor at a high, fast-changing voltage — a VFD output lead is the classic one — is separated from your signal wire by air and insulation. That is a capacitor, and a fast voltage change on one plate pushes current into the other. This is what a shield intercepts. The shield surrounds the pair, catches that current on itself, and drains it to ground before it ever reaches a conductor.
- Magnetic (inductive) coupling. Current in a nearby conductor produces a magnetic field, and that changing field induces a voltage in any loop it passes through. A thin foil or braid does very little about a low-frequency magnetic field. This is what the twist cancels. Twisting the pair makes each successive twist a small loop wound the opposite way from the one before it, so the induced voltages alternate in polarity and largely sum to zero over the run — and whatever is left lands on both conductors equally, which makes it common mode for the differential receiver to reject.
The Three Cable Shields You'll Actually See
Each cross-section is cut away on the right so you can see the layers. The percentages are typical figures for the construction, not a single published spec — braid coverage in particular varies by manufacturer and cable type.
Choosing Between Them
| Shield | Coverage | Strengths | Weaknesses | Where you'll see it |
|---|---|---|---|---|
| Aluminum/Mylar foil + drain wire | 100% — a continuous wrap, no gaps | Cheap, light, small outside diameter; very effective against high-frequency electric-field coupling precisely because there are no openings. | Thin, so relatively high resistance — a poor path for fault or low-frequency current. Work-hardens and can tear if the cable is flexed repeatedly. | Most instrument and multi-pair comm cable in fixed conduit or tray. |
| Braided tinned copper | Typically 60–95% — never 100%, because the weave has openings | Low DC resistance, so it handles fault current and works well at lower frequencies. Survives repeated flexing; easy to clamp to. | Bulkier, heavier, more expensive, and those weave openings let higher-frequency fields leak through. | Flexing and robotic cable, portable cordsets, low-frequency high-current areas. |
| Foil + braid (combination) | 100% from the foil, with the braid over it | Belt and suspenders: the foil gives complete coverage for high frequency, the braid gives the low-resistance path and the mechanical durability. | Largest, stiffest, and most expensive of the three. | High-noise plant runs, cable near VFD leads, industrial Ethernet. |
| Metallic conduit / armor (a rigid shield) | 100% if it is electrically continuous | Excellent shield and mechanical protection at the same time, with a heavy metal cross-section that makes a very low-impedance path. | Only works if every joint is bonded. A loose coupling, a painted joint, or a PVC section in the middle of the run breaks it electrically. | Plant-standard raceway for instrument and comm cable; armored cable (MC/ITC). |
How You Terminate the Shield Matters as Much as Having One
Same cable, same shield. The only difference is how many times it touches ground — and that difference decides whether the shield helps you or hurts you.
The Termination Rules
- Ground the shield at ONE end only — single-point (single-ended) grounding. Standard practice is to land it at the control panel / receiver end, where the plant's instrument ground bus is. Ground it at both ends and the ground potential difference between the two ends drives current through the shield — that is a ground loop, and now the shield is coupling noise into the pair instead of draining it away.
- Never use the shield as a signal conductor, and never as a substitute for signal common. It is a drain path, not a circuit conductor. Running signal return on the shield guarantees that noise current and signal current share the same piece of metal, which is exactly what shielding exists to prevent.
- Keep the shield continuous through every junction box. Splice drain to drain and keep that splice insulated from the enclosure, so the shield is one unbroken conductor from the field end to the single ground point. A shield landed on the box at every terminal strip is grounded in six places — six loops — and a shield left floating at a splice protects nothing past that box.
- Pigtail vs. 360° termination. A long twisted "pigtail" from the shield to a ground screw is a piece of wire, and a piece of wire is an inductor. Its impedance rises with frequency, so a pigtail gets progressively worse the higher the frequency of the noise you are trying to drain — and it can behave as a small antenna itself. Keep pigtails as short as you physically can, and for high-speed links such as industrial Ethernet use a 360° termination: a shield clamp or a metal connector backshell that bonds the shield around its entire circumference. At Ethernet frequencies that is not a refinement — it is the difference between a shield that works and a shield that doesn't.
You will occasionally see high-frequency installations that deliberately bond the shield at both ends, sometimes through a small capacitor at one end. The reasoning is that at radio frequencies a shield grounded at only one end can behave as an antenna, and the capacitor grounds it for high frequency while blocking the DC and 60 Hz current that would form the loop. For the low-frequency instrument and serial wiring in this course the rule you follow is single-point grounding at the panel end — but when a plant standard or a vendor manual tells you otherwise for a specific high-speed link, that is why.
One Note on Ethernet Cable Types
The same shielding vocabulary shows up on Ethernet cable, just under different letters. UTP is unshielded twisted pair — four twisted pairs, no shield, which is fine in an office and marginal in a plant. The FTP / STP designations add shielding: an overall foil around all four pairs, individually shielded pairs, or both. Industrial installs favor a shielded cable terminated in a shielded RJ-45 or M12 connector that makes a 360° bond to the shield. Note that both ends of an Ethernet link are commonly bonded, which is workable precisely because every Ethernet port is transformer-isolated — the signal pairs themselves have no DC path to the shield.
Wiring
Topologies — How These Are Physically Laid Out
Each standard implies a shape on the plant floor. Getting the shape wrong is one of the most common installation problems — especially hanging star branches off a bus that is supposed to be a straight trunk.
Vocabulary That Matters
Physical Layer vs. Protocol — Not the Same Thing
Technicians blur these constantly, and it causes real confusion on a troubleshooting call. Get this one distinction straight and a lot of the rest falls into place.
DeviceNet is the confusing one, because the name covers both halves: it specifies the protocol (CIP messaging over CAN) and the physical layer (a specific cable with a differential pair plus 24 VDC power). Same with PROFIBUS DP, which specifies both the protocol and RS-485 wiring. Modbus, by contrast, is purely a protocol and is happy on RS-232, RS-485, or Ethernet.
Industrial Network
Ethernet on the Plant Floor
Industrial Ethernet is electrically the same Ethernet as in the office. What makes it industrial is the installation practice, the connectors, the hardware ratings, and the protocols running on top.
The Physical Facts
- Topology: star. Every device gets its own cable back to a switch. Rings and redundant topologies exist with managed switches, but the basic building block is the star.
- Distance: 100 m (328 ft) per copper segment, device to switch. That is a hard number, and unlike RS-232's 50 ft it does not stretch by slowing down. Need more? Add a switch, or go fiber.
- Full duplex and switched: a modern switched network gives each device its own dedicated path, transmitting and receiving at the same time on separate pairs. Collisions are not a factor on a properly switched full-duplex network — the old CSMA/CD collision story belongs to shared hubs, which you should not be installing.
- Differential signaling: 10/100/1000BASE-T sends each pair differentially over twisted pair, so the exact noise-cancellation mechanism from the centerpiece page is doing the work here too. Use shielded cable in noisy areas.
- Transformer (magnetics) isolation at every port: the signal crosses a small isolation transformer between the RJ-45 and the transceiver chip, so there is no DC path between the two ends' electronics. Differential rejection plus galvanic isolation is what makes Ethernet notably tolerant of ground potential differences between panels.
- Fiber: for distance beyond 100 m, and for complete isolation when the ground offset is larger than the magnetics can handle or the two ends are on separate ground systems. Glass carries no current at all.
Protocols on Top
EtherNet/IP (the "IP" is Industrial Protocol, not Internet Protocol) is the CIP-based protocol common in the Rockwell / Allen-Bradley world. PROFINET is the Siemens-world equivalent. Modbus TCP is plain Modbus messaging wrapped in TCP/IP — simple, open, and everywhere.
Industrial Network
DeviceNet — A Device-Level Bus
DeviceNet sits lower in the plant than Ethernet. It's for connecting the small stuff — photoeyes, proximity switches, pushbutton stations, valve manifolds, motor starters, small drives — back to a controller over one cable, instead of home-running a pair of wires per device.
The Facts
- Built on CAN (Controller Area Network — the same bus in your truck), with the CIP protocol on top. CAN's differential signaling and built-in error detection are a large part of why DeviceNet is robust.
- Signaling: differential over a twisted pair (CAN_H / CAN_L). Same common-mode noise rejection as RS-485.
- Power in the cable: the standard DeviceNet cable carries 24 VDC on a second pair alongside the signal pair, so a sensor gets power and communication from one connector. That's the headline convenience of the bus.
- Topology: trunk and drop — a straight trunk with short drop cables branching off to devices. Individual drop length and total cumulative drop length are budgeted by the spec, not unlimited.
- Nodes: up to 64 (MAC IDs 0–63), and the master counts as one of them.
- Termination: a 121 Ω resistor at each end of the trunk — two total. Note it's 121 Ω, not 120: a 1%-tolerance value called out by the standard. Missing or extra terminators are the classic DeviceNet fault.
- Speed vs. distance (thick trunk cable): 500 kbps → 100 m, 250 kbps → 250 m, 125 kbps → 500 m. Every node on the network must be set to the same baud rate.
Reference
Side-by-Side Comparison
| Standard | Signaling | Conductors | Max Distance | Max Nodes | Typical Use |
|---|---|---|---|---|---|
| RS-232 | Single-ended, ground-referenced (±3 to ±15 V) | 1 signal wire per direction + shared ground | ~50 ft / 15 m (practical; the spec limits capacitance, not length) | 2 — point-to-point | Local programming ports, in-cabinet links, modems, printers |
| RS-422 | Differential (±200 mV receiver threshold) | Twisted pair per direction (typically 2 pairs) | ~4000 ft / 1200 m at low data rate | 1 driver, up to 10 receivers | One-to-many broadcast, long point-to-point runs |
| RS-485 | Differential (±200 mV threshold; −7 to +12 V common mode) | 1 twisted pair (2-wire) or 2 pairs (4-wire) + common | ~4000 ft / 1200 m at low data rate; much less at high speed | 32 unit loads (more with fractional-load transceivers or repeaters) | Modbus RTU, PROFIBUS DP, meters, drives, multidrop instrument buses |
| Ethernet (copper) | Differential, transformer-isolated, switched, full duplex | 4 twisted pairs (UTP/STP), RJ-45 or M12 | 100 m / 328 ft per segment (hard limit) | Practically unlimited via switches (bounded by addressing/bandwidth) | EtherNet/IP, PROFINET, Modbus TCP — controllers, HMIs, drives, I/O |
| DeviceNet | Differential CAN pair + 24 VDC power pair | 1 signal pair + 1 power pair in one cable | 100 m @ 500 kbps · 250 m @ 250 kbps · 500 m @ 125 kbps | 64 nodes (MAC ID 0–63, master included) | Device-level bus: sensors, actuators, starters, valve manifolds |
Troubleshooting
What Actually Breaks These
Comm faults are rarely exotic. In practice you'll find the same handful of things over and over — and most of them are installation errors, not component failures.
Termination Problems
- Missing terminator. The signal reaches the end of the trunk, finds no matching impedance, and reflects back down the line into the data behind it. Symptom: works fine at low baud or on a short run, falls apart when you speed up or extend it.
- Too many terminators. Somebody enabled the built-in termination jumper on a device in the middle of the bus. Now the driver is loaded by three or four resistors in parallel and the signal amplitude collapses. Exactly two, one at each physical end of the trunk.
- Field check: power down the bus, disconnect the master, and measure resistance across the pair. Two 120 Ω terminators in parallel read about 60 Ω. Around 120 Ω means one is missing; around 40 Ω (three) or 30 Ω (four) means someone terminated a mid-bus device.
Wiring and Grounding
- A and B (D+ / D−) swapped. The link is silent or garbled. Vendors label these inconsistently, so on a new install, swapping the pair at one device is a legitimate quick test. Be systematic: swap one, test, swap back if it doesn't help.
- Shield grounded at both ends. This creates a ground loop — the shield carries current between two grounds at different potentials, and that current couples noise into the very pair the shield was supposed to protect. Terminate the shield at ONE end only, normally the control panel end, and carry it through isolated at every junction.
- No signal common / reference. RS-485's common-mode range (−7 to +12 V) is generous but finite. Devices with no common reference can drift outside that window and stop communicating even though the pair itself looks perfect.
- Comm cable in the same tray as VFD output leads. This is the big one. Drive output cable is the noisiest thing in the building. Separate tray or conduit, and if it must cross power, cross at 90° — never run parallel.
Configuration and Limits
- Mismatched baud rate, parity, stop bits, or data bits. Every device on a serial bus must agree. One device set to 9600-8-N-1 on a 19200-8-E-1 bus can garble the whole segment whenever it transmits.
- Duplicate node addresses. Two devices answering to the same address produce intermittent, maddening behavior that looks exactly like a bad cable.
- No bias resistors. Between messages nobody drives the pair, so it floats, picks up noise, and receivers see phantom characters. Symptom: framing errors and junk bytes when the bus should be quiet.
- Exceeding distance — or exceeding it at that speed. 1200 m of RS-485 is a low-baud number; the same cable at 1 Mbps will not work. If a long link is marginal, dropping the baud rate is a valid diagnostic and often a valid fix.
- Star branches on a bus. RS-485 and DeviceNet want a single trunk with short stubs. Long branches off the middle behave like unterminated ends and reflect.
Separate TX and RX, or One Shared Pair? It Changes What a Symptom Means
Before you chase a one-direction failure, ask a question most technicians skip: does this link have a separate path for each direction, or does one pair carry both? The answer decides whether “it receives but it won't transmit” is pointing you at a wire or at a configuration.
| Standard | How the two directions travel | What “receives but won't transmit” means |
|---|---|---|
| RS-232 | SEPARATE conductors. TX on one pin, RX on another (pins 2 and 3 on a DB-9), plus signal ground. Full duplex. | A real, useful symptom. Half the link can die while the other half keeps working — suspect that one direction's conductor, pin, connector, or driver. |
| Ethernet (10/100BASE-T) | SEPARATE pairs. One transmit pair, one receive pair. Full duplex. (1000BASE-T uses all four pairs bidirectionally, so this reasoning applies cleanly to 10/100 only.) | A real, useful symptom. One broken pair, one bad pin in the jack, or one dead transceiver — link may even come up while one direction is dead. |
| RS-485, 2-wire | ONE shared differential pair for both directions. Half duplex — devices take turns. (4-wire RS-485 and RS-422 do have separate transmit and receive pairs; the reasoning here is for the common 2-wire case.) | Not a broken wire. If anything is getting through in either direction, the pair itself is intact — look at configuration and transmit enable instead. |
| DeviceNet / CAN | ONE shared differential pair (CAN_H and CAN_L) for both directions, all nodes. | Not a broken wire. Same logic — traffic in either direction proves the pair is good; the fault is at the node. |
On a separate-path link (RS-232, 10/100 Ethernet), that symptom is genuinely meaningful. One direction has its own wire, its own pin, and its own driver, so half the link really can fail while the other half keeps running. Go look at the hardware for that direction.
On a shared-pair link (2-wire RS-485, DeviceNet), it means something else entirely: if any traffic is getting through in either direction, the pair itself is intact. There is no separate transmit wire to be broken. So stop meggering the cable and look at the node — a wrong or duplicate node address, a baud rate or parity mismatch, a transmitter whose driver-enable (RTS / DE timing) never asserts, a device left in listen-only or monitor mode, or a transceiver whose driver half has failed while its receiver half still works.
Review