Start LAN Transformer Selection With the PHY, Not the Part Number
LAN transformer selection works best in one direction: from the PHY's requirements outward to the component. Picked the other way round, teams end up with a favourite part number and a portfolio of boards that each behave slightly differently at the margins. The PHY datasheet defines what the magnetic interface has to provide, and almost every disagreement about whether a given transformer will work can be settled there.
In practice four parameters decide almost everything, and they are interdependent rather than independent: turns ratio, open circuit inductance, isolation rating and common-mode behaviour. Reading any one of them without the other three is how datasheet comparisons go wrong. If you need to refresh what the magnetic block actually does between PHY and cable before reading further, our background piece on what a LAN transformer does sets the scene.
Turns Ratio: Matching the PHY and the Line
Most Ethernet magnetic channels are specified as a centre-tapped one-to-one ratio on each differential pair, because the PHY-side driver and the line are designed to be impedance matched through that unity relationship. Where a PHY deviates, it will say so explicitly, sometimes recommending a different ratio to accommodate a particular drive level or receiver sensitivity. Deviating from the PHY's stated recommendation without a reason is a redesign risk with little upside.
| Parameter | What it controls | Typical specification approach | Common mistake |
|---|---|---|---|
| Turns ratio | Signal amplitude ratio between PHY and line | Per differential pair, centre tapped | Choosing away from the PHY recommendation without reason |
| Open circuit inductance | Low-frequency response and droop | Minimum value at a stated DC bias | Comparing figures measured under different test conditions |
| Isolation rating | Safety withstand between port and system | Withstand voltage for a stated duration | Equating test voltage with continuous working voltage |
| CMRR / common-mode rejection | Susceptibility to common-mode noise | Attenuation in dB versus frequency | Comparing dB values quoted at different frequencies |
Also check how the centre taps are presented. In a discrete part they are separate pins with their own routing and termination requirements, and that termination, not the transformer core, frequently explains common-mode behaviour differences between two otherwise similar designs.
Why Open Circuit Inductance Sets the Low-Frequency Limit
Open circuit inductance, usually abbreviated OCL, describes how much magnetising inductance the winding presents, and therefore how the low-frequency end of the signal is treated. Too little and the waveform droops, which shows up as baseline wander rather than as obvious noise. Datasheets commonly quote a minimum figure in the hundreds of microhenries measured under a defined DC bias current.
The test condition is the part people skip. An inductance figure quoted without the bias current used to measure it is not comparable across datasheets, because magnetising inductance falls as bias current rises. When comparing candidates, put the test conditions side by side first and only then compare the numbers. This single habit resolves more apparent contradictions than any other comparison technique.
Isolation: What the Voltage Rating Does and Does Not Cover
Galvanic isolation is the reason the transformer exists at all in a twisted-pair port, separating the system ground from whatever is connected to the far end of possibly hundreds of metres of cable. The headline number is a withstand rating: the voltage the part is tested to survive for a defined duration, commonly quoted in the region of 1500 Vrms for Ethernet ports. That number is a test result, not an operating condition.
- Treat the withstand figure as type-test evidence, not as a voltage the design may apply continuously
- Confirm the working voltage expectation for your application separately from the withstand rating
- Remember that clearance and creepage on your own board matter as much as the component rating
- Keep the isolation boundary clean in layout, including any circuitry that crosses it
Isolation also carries a system-level implication for the products you ship. If the product is sold into a market requiring a specific safety file, the transformer's rating has to sit inside that file, which means choosing the part early enough that certification is not the surprise at the end of the project.
LAN Transformer CMRR and Common-Mode Noise
LAN transformer CMRR describes how effectively the magnetic channel rejects signals that appear identically on both conductors of the pair, as opposed to the differential signal carrying data. The common-mode choke integrated into the magnetic assembly does most of this work, presenting high impedance to common-mode current while leaving differential signals largely untouched.
The reason to care is practical rather than theoretical. Common-mode energy arrives from switching supplies, from adjacent pairs, from ground potential differences on long runs and from external interference, and the receiver sees whatever the transformer fails to reject. Because rejection is frequency dependent, figures are meaningful only when compared at the same frequency, and the range that matters is the one occupied by the signalling you are running.
Where EMI behaviour becomes the binding constraint, treat the magnetic choice and the enclosure return path as one problem. Our shielded vs unshielded RJ45 discussion covers the connector side of that boundary, and the two decisions should be made in the same review rather than in separate meetings.
Insertion Loss, Return Loss and the Signal Budget
Beyond the headline parameters, the channel has a budget. Insertion loss consumes signal amplitude, return loss reflects how well the impedance stays matched through the magnetics, and crosstalk between channels sets how much energy leaks between adjacent ports. Each is specified across frequency on a competent datasheet, and each should be added to the contributions from your board traces, connector and cable before concluding there is margin.
Compare like with like here as well. Two datasheets may quote different frequency ranges or different reference impedances, and a figure that looks better in one column often turns out to be quoted more favourably rather than genuinely superior. Where a parameter is missing from a datasheet entirely, treat that as information.
PoE: DC Bias Changes the Selection Entirely
Once power is carried on the same pairs, the transformer sees DC current through its windings and centre taps. That bias shifts magnetising inductance downward, which affects exactly the low-frequency behaviour discussed above, and the effect grows with delivered power. A magnetic channel that measures comfortably at low power can behave differently once the design supplies the higher classes defined for modern PoE implementations.
So specify the magnetic channel for the highest PoE class the product will actually deliver, not the one in the first configuration shipped. Check the bias current the inductance figure assumes matches your delivered power, verify the centre tap current capability, and remember that current ultimately flows through contacts as well, which is why the RJ45 gold plating choice and the magnetics choice belong in the same specification. Our RJ45 connector selection material covers how data rate class and port construction interact on that side of the boundary.
Discrete Transformer or Integrated Magnetic RJ45
The structural question is whether the magnetics sit on the board as a discrete component or inside the connector as part of an integrated RJ45 jack. Integrated parts save board area and simplify routing, which is why they dominate switching equipment. Discrete transformers give more freedom over the magnetic parameters themselves and can be placed where routing and thermal conditions are more favourable.
| Consideration | Integrated magnetic RJ45 | Discrete LAN transformer |
|---|---|---|
| Board area | Lower total footprint | More area required |
| Parameter freedom | Limited to the jack family offered | Wider choice of magnetic characteristics |
| Routing simplicity | Pairs terminate at one location | Additional routing between connector and magnetics |
| Thermal placement | Fixed at panel | Can be positioned for better conditions |
| Replacement strategy | Change the jack assembly | Change one component |
There is no universal answer, and the decision is usually made at the product architecture level rather than during detail design. What matters for this article is that the Ethernet transformer parameters above still have to be read and checked regardless of which structure is chosen, because integration changes the packaging, not the physics.
LAN Transformer Selection Checklist
- Begin with the PHY datasheet recommendation for turns ratio and terminations
- Compare open circuit inductance only against figures measured at the same DC bias condition
- Separate the withstand rating from the working voltage requirement in your specification
- Compare common-mode rejection at the frequencies your signalling actually occupies
- Specify the magnetic channel for the highest PoE class the product will deliver, not the first one
- Add magnetic, trace, connector and cable contributions before deciding there is margin
VITALCONN supplies both discrete LAN transformer parts and integrated magnetics inside RJ45 assemblies, and publishes dimensional drawings so that footprint decisions can be checked before layout rather than after it. If you are working from a reference design, send the PHY part number and the target PoE class and the shortlist usually becomes obvious quickly. Our RJ45 pinout and wiring guide covers the connector side of the interface for teams verifying their pin mapping at the same time.