Back to Resources

Attenuation of a Signal: Causes, Measurement and Fixes

LYH Circuits Teamattenuation of a signal

Attenuation of a signal is the reduction in signal power as the wave travels through a cable, an optical fiber, air, or a circuit trace, and engineers report it in decibels, where every 3 dB of loss cuts the power in half. A path that loses 10 dB delivers about a tenth of what the transmitter launched.

So which number tells you whether a link will work: the impedance tolerance printed on the stack-up drawing, or the sum of losses along the whole path? Most design reviews answer the first and skip the second. A board can sit comfortably inside a nominal impedance band and still fail, because impedance describes how reflections behave at discontinuities, while attenuation describes how much energy survives the journey from driver to receiver. Treating attenuation as a budget you add up, across fiber, copper, free space, and PCB traces, is the only way to catch that gap before a prototype does it for you.

Table of Contents

<a id="what-attenuation-of-a-signal-means-in-decibels"></a>

What Attenuation of a Signal Means in Decibels

Picture a garden hose stretched across a yard. Water pressure at the far end is lower than at the tap, and it drops further the longer the hose runs and the rougher its inner wall is. A signal behaves the same way: the medium takes energy out of the wave, and the amount taken grows with distance and with the material's losses. Attenuation of a signal is precisely that reduction in signal power as the waveform travels through a medium, and it's usually expressed in decibels, a convention NIST's review of fiber attenuation follows when it reports optical loss as decibels per kilometer so engineers can estimate link budgets and repeater spacing.

Decibels turn loss into a scale you can add. Because the unit is logarithmic, 3 dB of loss followed by another 3 dB is 6 dB, so an entire path becomes one sum: fiber plus connectors plus splices plus margin. Power uses 10 log10(Pin / Pout); voltage or amplitude uses 20 log10(Vin / Vout).

Remaining signal for common loss values

| Loss (dB) | Power remaining | Amplitude remaining | |---|---|---| | 1 dB | 79.4% | 89.1% | | 3 dB | 50.1% | 70.8% | | 6 dB | 25.1% | 50.1% | | 10 dB | 10% | 31.6% | | 20 dB | 1% | 10% |

Amplitude survives better than power at the same dB figure, and that isn't a quirk. Power scales with the square of voltage, so losing a third of your voltage costs you more than half your power. Keep both columns in mind, because a receiver's sensitivity spec is usually quoted in power while the swing you see on a scope is voltage.

Reading a datasheet with this table in front of you takes one step. A cable quoting 3 dB per 100 m at your operating frequency is telling you half your power is gone after one 100 m run and three quarters after two. Compare that against what the receiver needs and the distance limit falls out.

The same scale shows how far materials engineering can move a medium. NIST's history notes that around 1960 glass-clad fibers lost about 1 dB per meter, roughly 1,000 dB/km, which suited medical imaging but nothing long-haul. Ultrapure fused silica reached near 20 dB/km by 1970, germanium-doped cores near 4 dB/km by 1972, and loss fell below 0.2 dB/km by 1979, close to the theoretical low-loss region for silica. Modern telecommunications fiber runs near the 1,550-nanometer window, where attenuation is generally below 0.2 dB/km and more than 95% of optical power remains after 1 kilometer.

Take the loss figures on your own path, add them in dB, and convert the total into a percentage before you decide it's acceptable.

<a id="why-signals-lose-strength-on-each-path"></a>

Why Signals Lose Strength on Each Path

Loss is never one mechanism. Energy leaves a channel through at least four routes: absorption converts it to heat, scattering sends it in directions the receiver never sees, spreading dilutes it over a growing area, and discontinuities reflect a slice back toward the source. Which route dominates depends on the medium and on the frequency you run through it.

Fiber makes the frequency dependence easy to see. The Microsoft Learn archive on transmission losses?redirectedfrom=MSDN) cites typical figures of about 2.9 dB/km near 850 nanometers, 0.3 dB/km near 1,300 nanometers, and roughly 0.15 dB/km near 1,550 nanometers, and attributes the difference to how glass interacts with each wavelength rather than to any change in cable construction.

Here's the picture of what a broken path costs.

<a id="one-dominant-mechanism-per-medium"></a>

One dominant mechanism per medium

In glass, absorption and Rayleigh scattering share the work, and longer wavelengths lose less, which is why the 1,310 and 1,550 nanometer windows carry metropolitan, long-haul, and submarine traffic. Tight bends and poor splices add their own penalty on top of the distributed loss.

In copper, skin effect crowds current toward the conductor surface as frequency rises, so effective resistance climbs with every doubling of rate. The laminate underneath absorbs energy in step with frequency, the same material behavior that makes capacitor performance shift at high frequency. Rough foil lengthens the path the current must travel, which is why two boards with identical stack-ups and different copper treatments don't measure the same.

In free space, the wavefront spreads over a sphere that grows with distance, so power density falls even in a perfect vacuum, and walls, foliage, rain, and multipath subtract more.

At every transition, connectors, vias, layer changes, and splices add lumps of loss on top of the steady per-meter figure. A via stub can resonate and eat a band of frequencies while the rest of the channel looks healthy.

Run this checklist on your own path before you buy anything.

  • Distance: if doubling the length more than doubles the loss, a second mechanism is active.
  • Frequency: compare at the Nyquist frequency of your data rate, never at DC.
  • Material: check dissipation factor and copper roughness at the frequency you'll run.
  • Discontinuities: count connectors, vias, and layer changes, then measure one of them, not the whole chain.

<a id="wireless-path-loss-and-how-to-reduce-it"></a>

Wireless Path Loss and How to Reduce It

Free space alone costs predictable energy, and you can calculate it. FSPL (dB) = 20 log10(d in km) + 20 log10(f in MHz) + 32.44. Put a 2.4 GHz link at 1 km and you get 20 log10(1) plus 20 log10(2400) plus 32.44, which comes to roughly 100 dB. Two rules follow from the logarithms: doubling the distance adds 6 dB, and doubling the frequency adds 6 dB, so a move from 2.4 GHz to 5 GHz costs you about the same as walking twice as far.

Real environments charge more than free space does. Walls, foliage, rain, and multipath absorb or scatter energy, so measured path loss usually overshoots the free-space figure, and the gap is your obstacle allowance.

Practical rule: if the available link margin is smaller than path loss plus an obstacle allowance, move the antenna or shorten the path. Turning up transmit power is usually capped by regulation and rarely fixes a geometry problem.

Levers that work: shorten the distance, raise the antenna to clear obstructions, keep the first Fresnel zone free of buildings and trees, drop to a lower band where range matters more than bandwidth, and use a directional antenna to concentrate energy toward the receiver. Confirm the result by comparing received level against transmitted power on a link test, then subtracting the free-space number to see what the environment is charging you.

<a id="how-to-measure-attenuation-accurately"></a>

How to Measure Attenuation Accurately

Measuring loss works like weighing luggage: the number only means something if the scale is zeroed first and you state which units you're reading. Settle the sign convention before you quote anything. Report attenuation as a positive value, loss = 10 log10(Pin / Pout) for power and 20 log10(Vin / Vout) for voltage, so a positive figure always reads as loss on your drawing. Network analyzers present S21 as a signed ratio, so a passive channel appears as a negative number; flip it once, at the point of measurement, and don't let the sign wander into a spec table where someone will misread it.

Four methods cover most of what you'll measure.

  1. Apply the formula. Convert every raw reading to dB before comparing anything, and state whether you're quoting power or voltage ratios. Mixing the two conventions doubles your reported loss.
  2. Optical fiber, by cutback. Measure output power over the full length with a source of known launch power, cut the fiber back to a short reference length, measure again, and divide the difference in dB by the length removed to get dB/km. The same cut-and-remeasure idea works on short patch cords and multi-kilometer spools alike, which is why it remains the reference method for fiber attenuation.
  3. PCB and cable channels, by VNA. Calibrate at the cable ends, connect the coupon or channel, and read S21 as insertion loss across the full band you care about rather than a single convenient frequency.
  4. Location, by TDR. A time-domain reflectometer fires a pulse and watches the echoes, identifying impedance discontinuities in time. That answers a different question from S21: TDR tells you where the path changes, while S21 tells you how much signal a given frequency lost.

Below a few megahertz, a substitution method can resolve very large losses: a 2024 study in the Journal of Electromagnetic Engineering and Science used an intermediate-frequency substitution setup covering 9 kHz to 10 MHz to measure attenuation values up to 100 dB, which suits parts that would vanish into the noise floor of a swept measurement.

<a id="what-each-method-cannot-tell-you"></a>

What each method cannot tell you

  • The analyzer's own floor: a VNA's isolation and crosstalk limit the smallest transmission it can resolve, so a very lossy channel can read as flat at the floor rather than flat by nature.
  • The calibration plane: where you calibrate decides what counts as the device under test. Move the plane and the reported loss moves with it, the way a bathroom scale reads differently on carpet than on tile.
  • Fixture de-embedding: launches, cables, and probes contribute their own loss. The signal integrity testing guide from Wevolver reports that calibration quality and de-embedding can materially change the result, and cites a study finding that a 100 °C temperature increase can create about a 1 dB difference at 50 GHz.
  • A single number: a board can pass nominal impedance tolerance and still fail a high-speed link through distributed conductor and dielectric loss, surface roughness, or one localized via-stub resonance. Roughness-driven loss also grows with frequency.

Write the spec you actually need: test coupon, calibration plane, frequency range, and acceptance limit. A supplier who quotes a bare dB figure without those four has told you nothing you can check.

Confirm a clean run by repeating it, comparing against a model, and checking that the curve moves smoothly with frequency. A healthy loss curve rises gradually, like road noise building with speed, never in sudden steps. Jumps usually mean a fixture problem, a bad connector, or a resonance you haven't accounted for. And because every element along the path draws from the same budget, a dB of measurement error here is a dB of margin you either wasted or never had.

<a id="building-and-confirming-a-loss-budget"></a>

Building and Confirming a Loss Budget

Say you're budgeting a 10 km single-mode link at 1,550 nm. Your transmitter launches a known power, your receiver needs a stated minimum to hold its bit error rate, and everything between the two spends that power down.

Example loss budget for a 10 km fiber link

| Element | Loss per unit | Quantity | |---|---|---| | Fiber at 1,550 nm | 0.2 dB/km | 10 km | | Connector pair | 0.5 dB | 4 | | Fusion splice | 0.1 dB | 2 | | Subtotal before margin | 4.2 dB | one path | | Safety margin | 3 dB | 1 | | Total with margin | 7.2 dB | one path |

Fiber spends 2 dB, the four connector pairs spend 2 dB, the two splices spend 0.2 dB. That's 4.2 dB before any safety factor, and 7.2 dB once you add 3 dB for temperature swing, aging, and the repair splice someone will add in year three. Every per-unit figure is a design input, not a universal constant: replace each one with a datasheet value taken at your wavelength or frequency, because a connector loss quoted at 850 nm tells you nothing about 1,550 nm.

The whole exercise reduces to one inequality: transmitter output minus total path loss must still exceed receiver sensitivity plus margin. When it doesn't, you change a lever, not a hope.

Gather the inputs before you add anything: transmitter power, receiver sensitivity, path length, number of transitions, operating frequency or wavelength, and the temperature range the hardware must survive.

Then confirm the budget in five steps.

  1. Collect at the operating point. Pull loss values at the exact frequency or wavelength from datasheets and measurements, never from a headline figure quoted at another band.
  2. Sum the whole path. Include connectors, vias, splices, package transitions, and the launch, because the receiver sees all of them.
  3. Compare and subtract. Put the total against the available power budget, then take off a margin sized for your temperature range and service life.
  4. Change one lever at a time. Shorten the run, cut a transition, move to a lower-loss laminate or smoother copper, or shift the operating window, then rerun the sum. Changing three things at once tells you nothing about which one helped.
  5. Specify the acceptance test. Name the method and the limit, such as insertion loss measured by VNA across a defined frequency range, and request a coupon so verification happens on copper rather than on a promise.

For volume programs, we build custom stack-ups and hold impedance control across production, with a dedicated account manager tracking each build through fabrication and test.

<a id="reducing-attenuation-in-cables-wireless-links-and-pcbs"></a>

Reducing Attenuation in Cables, Wireless Links and PCBs

Every path loses signal for a different reason, so each one has a different lever worth pulling first. Spending against the wrong term is how loss budgets get wasted.

Attenuation levers by signal path

| Signal path | Main lever | What it changes | |---|---|---| | Optical fiber | Operate near 1,550 nm | Lowers dB/km loss | | Copper cable | Shorten run and size conductor | Lowers resistive loss | | Wireless link | Shorten distance and improve line of sight | Lowers path loss | | PCB trace | Low-loss laminate at high frequency | Lowers dielectric loss | | PCB copper | Smoother copper foil | Lowers conductor loss at high frequency | | Vias and connectors | Backdrill stubs and reduce transitions | Lowers discontinuity loss | | Operating temperature | Keep the operating range low | Limits loss increase |

The rows overlap on purpose. A 28 GHz backplane might need the laminate, the foil, and the backdrilling rows all at once, because the budget cares about the sum, not which row contributed it.

<a id="when-premium-material-earns-its-price"></a>

When premium material earns its price

At high frequency, current crowds toward the conductor surface, so surface roughness acts like extra resistance. Smoother foil lowers insertion loss, and the advantage grows as frequency climbs. The per-inch gain is small, which is exactly why length decides whether it matters: across a 3 inch board the difference barely registers, while across a 20 inch backplane at 28 GHz the same per-inch gap adds up to a real slice of the budget.

That arithmetic settles most laminate arguments. A short interconnect usually has margin to spare. A long backplane with several vias, two connectors, and a hot enclosure can consume every dB you thought you had, which is when low dissipation factor and smooth foil stop being datasheet decoration. Temperature plays a role, since conductor resistance rises with heat: a board that meets its budget on the bench can miss it inside a warm chassis.

Compare three numbers, not two: trace length, Nyquist frequency, and the allowable insertion loss left in your budget after connectors and vias. Material cost only makes sense against the third one.

Our high-frequency boards use Rogers, PTFE, and high-speed materials for RF, microwave, 5G, and mmWave work, selected by Dk and Df and verified with SET2DIL, Delta-L, and VNA testing. If your stack-up calls for a specific laminate, the Rogers PCB manufacturing process matters as much as the material name, because press cycles, bake profiles, and drilling parameters shift the loss you measure on the finished board.

For a visual walkthrough of how these losses show up on a real channel, this segment is worth ten minutes.

<iframe width="100%" style="aspect-ratio: 16 / 9;" src="https://www.youtube.com/embed/pBEcZI-P2Lc" frameborder="0" allow="autoplay; encrypted-media" allowfullscreen></iframe>

Backdrilling is the cheapest dB in many designs. Removing an unused via stub kills a resonance that no laminate change can reach, and it costs a drill cycle rather than a material upgrade.

<a id="misconceptions-and-hard-physical-limits"></a>

Misconceptions and Hard Physical Limits

Three beliefs account for most of the bad decisions that show up in a signal integrity design review.

  • "More transmit power fixes it." It doesn't. Cranking the driver raises noise along with signal, heats the part, and runs into regulatory or device limits, while the receiver's sensitivity floor stays exactly where it was. Geometry and material change the physics; power only buys headroom you may not be allowed to use.
  • "Controlled impedance means low loss." Impedance and attenuation are different measurements. A trace can hit its tolerance target and still lose too much energy to distributed conductor and dielectric loss, or to one stub resonating in the middle of your band.
  • "A repeater removes the loss." Amplifiers and regenerators restore amplitude by adding noise, cost, power, and a failure point. They widen the budget; they don't refund it.

The physical ceiling is what you plan against. Silica's low-loss region bottoms out near 0.2 dB/km, and standard single-mode fiber in the 1,550 nanometer window operates close to that floor already, which is why long-haul budgets are so sensitive to connector count rather than fiber quality.

Hollow-core fiber is now pushing past that ceiling. Microsoft reported in 2025 that a hollow-core optical waveguide reached 0.091 dB/km measured loss at 1,550 nm and stayed below 0.2 dB/km across a 66 THz window. That result comes from guiding light through air rather than glass, so it says nothing about the laminate under your BGA. Design to the actual limit of the medium you're using, not to a record set by a different structure.


Attenuation is a sum, not a headline, and the fastest way to close a stubborn link budget is to measure the parts of it separately. Lianyuhong Technology Co.,Ltd builds high-frequency and high-speed PCBs on Rogers, PTFE, and hybrid stack-ups, and we verify insertion loss with SET2DIL, Delta-L, and VNA testing so the number on the datasheet matches the number on your board. If you're chasing a few dB on a backplane or an RF front end, send us your stack-up and loss target, and visit Lianyuhong Technology Co.,Ltd to start the conversation.