A Customer Burned a Standard Transformer in a PoE Design. Here’s What They Should Have Known.

I get this question more often than I’d like: “We used a standard network transformer and the PoE link works fine in the lab, but fails after an hour under load. What’s wrong?” Almost every time, the answer is the same. They used a standard transformer in a PoE circuit. Let me explain exactly what goes wrong — and what to specify instead. Why standard transformers fail under PoE A standard network transformer is designed for one job: transferring AC differential signals from a PHY chip to a cable, with galvanic isolation in between. The core material, winding geometry, and inductance are all optimized for AC operation with no DC component. PoE introduces DC current through the transformer windings. A PoE switch (PSE) pushes anywhere from 350mA to 960mA of DC current through the cable pairs to deliver power. That DC current flows through the transformer windings. In magnetics, DC current through a winding creates DC magnetic flux in the core. As DC flux builds up, it consumes the core’s available flux range. When you add AC signal flux on top of a partially saturated core, the core can’t respond fully to the AC component. Inductance drops. OCL collapses. The transformer that measured 1000µH in the lab now behaves as if it has 200µH at operating temperature under full PoE current. Your 1000BASE-T link degrades and eventually drops. The frustrating part: it often works in the lab. Benchtop PoE testing usually runs at lower current than worst-case operation. The failure shows up in the field, at full load, at elevated temperature — exactly when you can’t easily debug it. What the standards actually require The IEEE 802.3 PoE family has grown significantly:

802.3af (Type 1): up to 350mA, 15.4W at the PSE 802.3at (Type 2): up to 600mA, 30W at the PSE 802.3bt Type 3: 600mA across two pair groups, 60W 802.3bt Type 4: 960mA across two pair groups, 90W

For 802.3bt designs, every transformer section (all four pairs) carries DC bias simultaneously. Your magnetics must handle full-load DC current on all four pairs while maintaining ≥ 1000µH OCL on each. What PoE-rated magnetics actually do differently The core difference is an air gap in the core geometry. An air-gapped core has higher magnetic reluctance — it takes more current to reach the same flux density. This pushes the saturation threshold to higher current levels, allowing the transformer to maintain its OCL spec even with full PoE bias flowing through the windings. PoE transformer datasheets should show OCL vs. DC bias current curves, or at minimum specify the OCL at rated DC bias. That’s the number that matters for your design — not the zero-bias OCL. DCR is more important than most engineers realize DCR (DC Resistance) of the transformer winding determines how much heat the transformer generates under PoE current. The math is P = I² × R, and at high PoE currents this adds up fast. At 802.3bt Type 4 (960mA per pair, four pairs, DCR 0.5Ω per winding): that’s roughly 1.85W dissipated in the transformer itself. In a small SMD package on a dense board, that heat matters. If your operating temperature is already 70°C ambient, a 1.85W source inside the transformer body can push its core temperature over the rated limit — which is another failure mode that only shows up in field conditions. Check DCR in the datasheet and calculate your worst-case thermal load before committing to a transformer. PSE or PD — same transformer spec Whether you’re building a PoE switch (PSE, sourcing power) or a PoE camera (PD, receiving power), the transformer sees the same DC bias current. The direction is different, but the magnitude is determined by the PoE class. The transformer specification is identical for PSE and PD at the same PoE class. The design rule I now follow Any design with a PoE controller — PSE or PD, regardless of class — gets a PoE-rated transformer. The cost difference is small. The risk of not using one is a product that fails in field conditions you didn’t test for. For my recent designs, I’ve been sourcing PoE-rated Gigabit magnetics from Voohu Technology (www.voohuele.com). They stock 802.3af, 802.3at, and 802.3bt rated parts, provide OCL-under-bias specs in their datasheets, and ship from 50pcs at DHL speed. When I’m doing a new design, I send them the PoE class and PHY chip, and they match the part.