Conductor Loss vs Dielectric Loss: Computing Line Attenuation
Total attenuation is α = αc + αd, converted from Np/m to dB/m by multiplying by 8.686. Conductor loss rises as √f and dielectric loss as f. A line filled with εr = 4.4 and tanδ = 0.02 already loses 3.82 dB/m in the dielectric at 1 GHz.
Engineer’s note: Below about 3 GHz the main loss is in the copper, through skin effect. Past 10 GHz, dielectric loss from the laminate's Df climbs fast. High frequency doesn't automatically mean an expensive PTFE board: copper roughness adds a lot of conductor loss, and switching to VLP (very low profile) copper can be the more cost-effective fix.
For practitioners
A transmission line loses power in two places. Current flowing in the metal heats the conductors. The alternating field heats the dielectric between them. You can estimate each loss separately, put both in the same units and add them. Comparing the two terms then tells you which one to attack: the copper or the laminate.
Two loss mechanisms, one exponent
On a low-loss line the voltage wave decays as e^(−αz), where α is the attenuation constant in nepers per metre. Both loss mechanisms add to the same exponent, so the total is just their sum:
Engineers work in decibels, so convert the result. Power decays as e^(−2αz), and 10·log10(e²) = 20·log10(e) = 8.686:
You can add the two terms in Np/m and convert once, or convert each one and add in dB/m. Both give the same answer, because the conversion is linear. The one thing you cannot do is add a value in nepers to a value in decibels.
The calculation therefore has two independent branches. One branch starts from frequency and conductor conductivity, finds the surface resistance and then αc. The other starts from frequency and the dielectric properties and finds αd. Each branch is scaled by 8.686, and the two results are summed.
Conductor loss from surface resistance
At RF the current crowds into a thin layer at the conductor surface. The skin depth is δs = √(2/(ωμ0σ)), and the conductor behaves like a sheet with surface resistance Rs = 1/(σδs):
For copper (σ = 5.8e7 S/m) at 1 GHz, Rs = 8.25 mΩ, and the skin depth is 1/(σRs) ≈ 2.09 µm. Rs scales with √f, so it is about 2.61 mΩ at 100 MHz and 26.1 mΩ at 10 GHz.
For microstrip, Pozar gives a simple approximation that treats the strip as wide:
Here W is the strip width in metres. For a 50 Ω copper line at 1 GHz this becomes αc ≈ 1.65e-4/W Np/m. With W in millimetres that is αc ≈ 1.43/W(mm) dB/m. Narrow strips lose more, and lower-impedance lines lose less for the same width. Because W and Z0 are linked by the stackup, the impedance and width you get from the microstrip width calculation feed directly into this estimate.
Dielectric loss and the filling factor
For a low-loss TEM line completely filled with a non-magnetic dielectric, such as stripline or filled coax, the dielectric attenuation depends only on the material and the frequency:
The line geometry does not appear. Every field line passes through the same lossy material, so impedance and width do not change αd.
Microstrip is different, because part of its field is in the air. Pozar's expression includes a filling factor:
As a check, set εeff = εr. The (εeff − 1)/(εr − 1) term becomes 1, and the expression reduces to k0·√εr·tanδ/2, which is the filled-line result. For any real microstrip εeff < εr, so its dielectric loss on a given laminate is somewhat lower than a stripline on the same material.
Worked example: dielectric loss at 1 GHz
Take a TEM line completely filled with εr = 4.4, tanδ = 0.02, at 1 GHz.
- Free-space wavenumber: k0 = 2πf/c = 20.96 rad/m.
- In the dielectric: k = k0·√εr = 20.96 × 2.098 = 43.96 rad/m.
- Attenuation: αd = k·tanδ/2 = 43.96 × 0.02 / 2 = 0.440 Np/m.
- In decibels: 0.440 × 8.686 = 3.82 dB/m.
That is 0.382 dB per 10 cm, from the laminate alone and before any copper loss. For the total, compute αc for the actual conductor geometry, convert it with the same 8.686 factor, and add.
Which loss dominates as frequency rises
The two terms scale differently. With smooth conductors in the skin-effect regime, αc rises as √f through Rs. If tanδ stays roughly constant, αd rises as f, because k is proportional to f. On a log-log plot the dielectric loss is a line of slope 1 and the conductor loss is a line of slope 1/2.
For the filled line above, αd is 0.382 dB/m at 0.1 GHz, 3.82 dB/m at 1 GHz and 38.2 dB/m at 10 GHz, so it rises tenfold per decade. Over the same decade the conductor loss rises by only √10 ≈ 3.16. Lines with different slopes must cross. Below the crossover, conductor loss dominates. Above it, dielectric loss dominates, and its share keeps growing.
If you know both terms at one reference frequency f0, the crossover follows directly:
At the crossover each term carries half the total loss. The exact crossover point depends on Rs, Z0 and W, so the conductor line in the figure is illustrative. The slopes are not illustrative. They follow from the physics.
This tells you where to spend effort. At low frequency, wider traces, lower-impedance lines or better conductors help most. At high frequency, a lower-tanδ material is the only change that moves the dominant term.
In plain words · for students and beginners
Line attenuation in plain words
An everyday picture
Think of a garden hose that leaks in two ways. Water rubs against the walls of the hose and loses energy. That friction is like conductor loss, the heat made by current flowing in the metal. The hose is also filled with a sponge that soaks up a little water as it passes. That is like dielectric loss, the heat made in the insulating material (the dielectric) between the conductors. The total loss is just the two leaks added together.
The picture breaks down in two places. RF current does not use the whole metal. It flows in a very thin layer at the surface, called the skin depth, which is about 2 µm in copper at 1 GHz. Also, the "sponge" is not soaking anything up. The insulating material cannot keep up perfectly with the field as it flips back and forth billions of times per second, and that lag turns some energy into heat.
The same example, step by step
Each loss is worked out on its own path, put into decibels, and then added.
For the dielectric part, use a line fully filled with a material whose εr (how strongly it stores electric field) is 4.4 and whose tanδ (how lossy it is) is 0.02, at 1 GHz:
- Find how fast the wave's phase turns in free space: 20.96 radians per metre.
- Multiply by √4.4 = 2.098 to get the value inside the material: 43.96 radians per metre.
- Multiply by tanδ and halve it: 43.96 × 0.02 / 2 = 0.440. The unit is nepers per metre, a natural-log unit of loss.
- Multiply by 8.686 to get decibels: 3.82 dB per metre.
For the metal part, start with the surface resistance Rs, which is the resistance of the thin skin layer. For copper at 1 GHz it is 8.25 mΩ. Divide it by the line impedance and the strip width, then multiply by 8.686. Add the two dB values to get the total.
Why the answer changes with frequency
If you go ten times higher in frequency, the dielectric loss becomes ten times larger, so 3.82 dB/m becomes 38.2 dB/m. The metal loss grows only about 3.16 times, because it follows the square root of frequency. At low frequency the metal usually loses more. At high frequency the dielectric wins.
Mistakes beginners make
- Adding nepers to decibels. Put both numbers in the same unit first.
- Using 4.343 instead of 8.686. 4.343 converts power loss in nepers. For α, the right factor is 8.686.
- Using εr where εeff belongs. On microstrip, part of the field is in air, so the formula needs εeff, the "average" permittivity the wave actually sees.
- Assuming loss is fixed. A loss value is only good at the frequency where you calculated it.
Remember
- Total loss = conductor loss + dielectric loss, in the same units, and × 8.686 turns Np/m into dB/m.
- Conductor loss grows with √f. Dielectric loss grows with f.
- At high frequency, a lower-loss board material helps more than wider copper.
Related pages on RF View
Frequently asked questions
How do I convert attenuation from Np/m to dB/m?
Multiply by 8.686, which is 20·log10(e). For example, 0.440 Np/m is 3.82 dB/m. Convert both loss terms the same way before you add them, or add them in Np/m and convert once.
Does conductor loss or dielectric loss dominate at high frequency?
With smooth conductors in the skin-effect regime, conductor loss grows with √f and dielectric loss grows with f if tanδ stays roughly constant. Above some crossover frequency the dielectric term is therefore the larger one, and the gap keeps widening as frequency rises.
What is the surface resistance of copper at 1 GHz?
Rs = √(ωμ0/(2σ)), which for copper with σ = 5.8e7 S/m at 1 GHz gives 8.25 mΩ. It scales with √f, so it is about 26.1 mΩ at 10 GHz.
Why is microstrip dielectric loss lower than stripline loss on the same laminate?
Part of the microstrip field runs in the air above the board, so only part of the energy sees the lossy dielectric. Pozar's microstrip formula captures this with a filling factor based on εeff. It reduces to the fully filled result k·tanδ/2 when εeff equals εr.