Loss At The Nyquist Frequency: Why 25 Gbps Is Specified At 12.5 GHz

Interface specifications are often written in terms of a data rate, and the loss limit is applied at half that number. The reason is that a non return to zero signal is a series of symbol transitions, and the fastest pattern it can contain alternates every symbol, producing a fundamental at half the bit rate. That is the Nyquist frequency, and it is where the channel loss limit is quoted.

This article explains why the specification is written that way, what the second and third harmonics add, and how to use the figure when choosing a laminate and planning a route.

The practical value of understanding the convention is that it allows a loss budget to be built from a single number in a specification rather than from a full channel analysis.

Where The Half Rate Comes From

A bit stream at a given rate carries its energy across a band that extends well above the symbol rate, because of the edges. The highest frequency at which the data itself has content is the alternation pattern, one zero one zero, which repeats at half the bit rate. Everything above that is a consequence of the edge shape rather than of the data pattern.

The receiver has to resolve that alternation, so the loss at that frequency is the loss that determines whether the eye is open. Measuring a channel above the Nyquist frequency is still useful, because a very fast edge has content there and a channel that falls away sharply above it will distort the edges even if the eye at the sampling instant is acceptable.

<img src="https://www.gopcba.com/wp-content/uploads/2026/08/pcb15.jpg" alt="Insertion loss curve with the Nyquist frequency marked” />

Why Not Just Quote A Frequency

Quoting a data rate rather than a frequency is a convention that keeps the specification close to the interface it describes. It also allows the same specification to be applied to a link whose coding scheme differs, since the relationship between the bit rate and the required bandwidth depends on the encoding.

Where a link uses a coding scheme that changes the spectral content, the effective Nyquist frequency moves. A scheme that runs at a higher symbol rate to carry the same payload, or one that shapes the spectrum deliberately, changes the frequency at which the loss limit applies, and the specification will state the relationship.

It also explains why a channel can look acceptable in a low frequency measurement and fail in service. A sweep that stops below the Nyquist frequency reports the easy part of the response and omits the region where the material and the discontinuities do their damage.

Building A Loss Budget From The Number

The specification gives an allowed loss at a frequency. The budget divides that allowance between the board, the package, the connector and the cable. The board’s share is what the laminate and the geometry have to deliver, and it should be estimated before the stackup is chosen rather than after the layout is complete.

The estimate needs the loss per unit length of the chosen material at that frequency, which the supplier publishes, and the length of the route. Adding the via and connector contributions produces a total that can be compared with the allowance. Where the total exceeds it, the options are a lower loss material, a shorter route or a lower data rate, and all three are cheapest to evaluate on paper.

Equalisation And What It Buys

Most fast interfaces include equalisation at the transmitter, at the receiver or both. The equaliser compensates for the predictable part of the channel response, which extends the reach of a given material considerably. What it cannot do is recover information that the channel has removed, and it cannot correct a channel whose response does not match the profile the equaliser expects.

The practical consequence is that equalisation shifts the boundary rather than removing it. A channel that is slightly over the loss limit may work with equalisation, and the same channel with an unexpected resonance or a bad connector may not, even though its loss figure is better. The general practice for high frequency trace routing applies to both.

The third and fifth harmonics matter for a different reason. They determine the shape of the edge rather than the ability to resolve a symbol, and a channel that attenuates them heavily rounds the transitions, which reduces the timing margin even when the amplitude at the sampling instant is adequate. This is why two channels with the same loss at the Nyquist frequency can differ in measured performance, and it is why some specifications place a limit at a higher frequency as well.

Choosing Material Against The Number

With the allowed loss and the route length known, the required loss per unit length follows directly, and the material can be selected against it rather than by reputation. The figure to compare is the insertion loss per unit length at the Nyquist frequency, usually quoted by the supplier for a standard test structure.

Comparison is easier when the figures come from the same measurement method, because the dieletric and the conductor contributions are separated differently by different suppliers. Where the figures are close to the requirement, the choice should be made with the low loss laminate selection criteria in mind rather than on loss alone.

Routing To The Number

Once the material is chosen, the layout has a length budget. Every additional millimetre of a controlled impedance route consumes part of it, and so does every via and every connector. Plotting the route against the budget during layout keeps the channel within its allowance, rather than discovering at the end that the routing is acceptable and the loss is not.

Where the budget is tight, the geometry offers a small lever. A wider trace on a thicker dielectric reduces conductor loss at the same impedance, and the trade with dielectric loss usually favours the wider geometry on a long, low loss channel. The optimum is found with a field solver rather than by rule of thumb.

Why The Convention Matters In Practice

The half rate convention is the reason two engineers can discuss a channel with a single figure and mean the same thing. It is also the reason a data sheet that quotes loss at the bit rate rather than at the Nyquist frequency is misleading, and one that quotes it at both is unusually helpful.

When a specification is ambiguous, the safe assumption is the one that gives the larger loss allowance to the channel, because the margin then covers the interpretation as well as the design. Confirming the convention with the interface specification before the stackup is frozen costs nothing.

Temperature is a variable that is easy to leave out. The loss tangent of most laminates rises with temperature, so a channel that meets its budget at room temperature may not meet it at the upper limit of the operating range. Where the margin is small, the budget should be evaluated at the high temperature end, and a material with a flatter loss curve over temperature may be worth the additional cost.

FAQ

Is loss at the Nyquist frequency enough to describe a channel? It is a useful summary, but a channel with the same loss at that frequency and a much worse response above it will perform differently. Where the specification allows, the loss at one and a half times the Nyquist frequency is a better indicator of edge quality.

Does a higher data rate always need a better material? Not if the route is shorter. A short channel on a standard laminate can meet a specification that a long channel on a low loss material cannot, because loss is a product of the material and the length.

Should the connector be included in the board budget? The connector is part of the channel and its loss counts against the allowance. Where the connector is chosen later, its share should be reserved rather than discovered.