Industrial real-time Ethernet is a family of protocols that run control traffic over standard Ethernet hardware while bounding when each frame arrives, not just whether it arrives. PROFINET, EtherCAT, POWERLINK and EtherNet/IP each solve that timing problem differently. TSN is the IEEE layer that lets them share one wire with everything else. AEI Corporation works at the level where those differences matter: protocol stack integration, driver and PHY bring-up, cycle-time budgeting, and pre-compliance testing before a device goes to a certification lab.
Determinism is a question of variance, not throughput. A gigabit link with an unbounded worst case is useless to a control loop; a 100 Mbit/s link with a guaranteed transmission slot is not. Standard switched Ethernet is non-deterministic because queuing delay depends on what else is in the queue. Add traffic and the tail of the latency distribution moves, even when average link utilisation is low.
Every protocol on this page is an answer to that one problem, and they answer it in one of three ways: schedule the medium (reserve a phase or a time slot so control frames never contend), change what a frame is (one frame serving many nodes instead of one frame per node), or police the traffic (admission control and per-stream shaping in the switch). Which answer suits a machine depends far more on its topology and node count than on its data rate.
| Protocol | IEC 61158 type | IEC 61784-2 CPF | Determinism mechanism | Node hardware |
|---|---|---|---|---|
| PROFINET IRT | Type 10 | CPF 3 | Hardware-reserved IRT phase; PTCP / IEEE 1588 sync | ASIC or FPGA required for IRT class C |
| EtherCAT | Type 12 | CPF 12 | On-the-fly frame processing; one telegram per segment | ESC ASIC per node |
| POWERLINK | Type 13 | CPF 13 | Software-scheduled isochronous phase; managing node polling | Standard Ethernet NIC; no specialist silicon |
| EtherNet/IP | Type 2 | CPF 2 | CIP implicit messaging; determinism via network topology | Standard Ethernet adapter |
Start from the constraint that actually binds. Axis count and cycle budget point toward EtherCAT or IRT-class PROFINET. Ecosystem and end-customer mandate frequently decide it before engineering does. A machine sold into a plant standardised on one protocol will speak that protocol. Certification cost and time is the third factor and the most often under-planned: the conformance path, not the stack, is usually what sets the schedule.
The honest answer for most device makers is that the protocol is chosen for them; the real question, covered in the child pages above, is what implementing it well costs.
We integrate and port protocol stacks onto customer hardware, do the driver and PHY-level work underneath them, budget cycle times against real topologies before hardware is committed, and run pre-compliance testing so a device arrives at an accredited test lab already passing. Where a design is being taken onto a converged network, we do the TSN configuration and verification work alongside it.