The LAN Transformer: PoE, Speed, and Selection

In Part 1 of From Zero to PoE, we covered the LAN transformer's basic job: signal coupling, DC blocking, common‑mode rejection, and isolation. Now we add two extra dimensions that turn a simple transformer into a critical design decision — PoE current and data rate.

1. How does PoE get added? If your system uses PoE, the transformer takes on an extra role.

The core engineering challenge in PoE is: how do you send data and power over the same pair of wires without them interfering?

The answer is simple — frequency multiplexing. Data is high-frequency AC, power is DC. They occupy different frequency bands, so they can coexist on the same physical medium. The question is how to inject the DC power without messing up the differential signal.

That injection point is the center tap.

In a PoE system, the PSE (power-sourcing equipment) side applies 48V DC through the center tap. The DC current flows into the center tap, splits into two paths through each half of the winding, and recombines on the secondary side to go out to the cable.

The key is: the DC currents in the two halves of the winding flow in opposite directions (assuming ideal symmetry), so the DC fluxes they create in the core cancel out — no net DC flux. That way, the DC power superimposes on the signal path without disturbing the differential signal or biasing the core.

On the PD (powered device) side, the transformer at that end does the reverse: it extracts the DC power from the center tap and feeds it to the downstream DC-DC converter.

That's the underlying principle of PoE — the center tap provides a DC path, the differential signal uses the winding ends, and the two coexist on the same core without interfering.
**
But there's a catch: **the transformer must be designed to handle that DC current. Not every transformer can do that.

5. What's the difference between a "PoE" transformer and a regular one?

A lot of engineers ask this. The answer is straightforward: it's about structural parameters, not a functional category.

When PoE current flows through the center tap, it creates a constant DC bias field in the core. That bias pushes the core's operating point toward saturation.

Every core has a physical limit — saturation flux density (Bsat). As you approach that limit, permeability drops sharply, and inductance plummets. When inductance drops, the differential signal can't couple properly, return loss worsens, and BER goes up. In severe cases, the link simply fails.

So PoE capability boils down to one question: under a given DC bias, can the transformer still maintain enough inductance?

That depends on three design parameters:

1. Wire gauge — current capacity; too thin and it overheats
2. Core cross-sectional area — larger area means more saturation margin
3. Core material — high Bs materials can handle higher bias In the datasheet, the difference between non-PoE and 720mA is not a software switch — it's a physical difference. A non-PoE transformer isn't "unsupported" — it's simply "not optimised for DC bias." Forcing PoE current through it might work temporarily, but in the long run — saturation, heat, signal collapse — it's only a matter of time.

The current ratings you see on transformer models — 350mA, 720mA, 900mA, 1200mA, 2000mA — correspond to these standard levels.

6. Why is data rate also a hard constraint?

Same core structure, but why can't you use a transformer rated for one speed with another?

Because core materials and winding structures have frequency responses.

• 10/100BASE-TX: bandwidth ~20 MHz, low-frequency losses are minor, more turns can be used
• 1000BASE-T: four pairs full‑duplex, requires ≥100 MHz bandwidth
• 2.5/5G BASE-T: needs ≥200 MHz
• 10G BASE-T: **needs ≥500 MHz Higher rates tighten the margins for **insertion loss and return loss. A transformer optimised for 100M might have excessive loss or reflections at high frequencies — the Gigabit link may fail to negotiate or may constantly downshift.

Conversely, using a Gigabit transformer on a 100M link usually works fine — bandwidth margin exists. But the reverse is not true.

High-frequency signals in a low‑frequency‑optimised winding will hit distributed capacitance and leakage inductance, causing signal integrity issues. This isn't "slightly worse performance" — it's "will it work at all."

7. Back to that black box

The LAN transformer is not just an "accessory" between the PHY and the RJ45.

Physically, it does one thing: find an engineering compromise between electrical isolation and signal integrity.

• No electrical connection between primary and secondary, but magnetic coupling — signal gets through, ground doesn't
• The center tap is the DC injection point — power is added, signal isn't disturbed
• The B-H curve of the core determines how much DC bias it can take — PoE capability is not a feature, it's physics
• The core's frequency response dictates the maximum data rate — speed grade is not a configuration, it's a structure
• Once you understand what's happening inside that black box, selection becomes a matter of physics:

How much DC current do you need to carry? How fast does the signal have to go? What isolation voltage does your safety requirement call for?

Answer those three questions, and that little black box won't be a black box anymore.

📌 Next up in From Zero to PoE: The transformer already has some common-mode rejection, so why do we need a dedicated common-mode choke? Part 3 breaks down their respective roles — the transformer handles isolation and coupling, the choke handles noise suppression — and explains how they complement each other.