L
Admin
Most of us type one Dk and one loss tangent from the laminate datasheet into the solver and move on. For narrowband filters that is acceptable. For broadband interconnects, TDR and channel simulation, it quietly produces non-causal S-parameters.
1. Why constant values are wrong
Real dielectric permittivity and loss are linked by the Kramers-Kronig relations. If the loss is non-zero, the permittivity must change with frequency. A model with frequency-independent Dk and tan delta violates this, so time-domain results show pre-cursor ringing or a step response that starts before the signal arrives.
2. The Djordjevic-Sarkar (wideband Debye) model
It describes permittivity as a smooth function with a nearly constant loss tangent over a chosen frequency span (for example 1 kHz to 1 THz). You give it one Dk and tan delta at a reference frequency, and it returns a causal curve where Dk falls slowly with frequency and loss stays close to constant. Most commercial solvers offer it as a built-in option; openEMS and other open tools can approximate it with several Debye or Lorentz poles.
3. Picking the reference point
- Laminate datasheets often quote Dk at 1 or 10 GHz, sometimes from different test methods (stripline resonator, split post, clamped stripline). The values can differ by several percent.
- Use the value measured closest to your band, from the same test method if you have it, and note which one you used.
- Remember that the design Dk recommended by the vendor for impedance calculation may differ from the process Dk used for QC.
4. Glass weave and anisotropy
Woven glass laminates are anisotropic: in-plane Dk differs from the z-axis Dk. Microstrip and stripline fields mostly see z-axis Dk, but coupled lines and filters with strong horizontal fields can see the in-plane value. Spread-glass styles reduce the fibre weave skew in differential pairs.
5. Combining with roughness
Dielectric and conductor models must be fitted together. If you tune roughness to match measured loss while using a non-causal dielectric, you may get the right insertion loss for the wrong reason and the phase will still disagree.
6. Quick sanity check
Run the extracted S-parameters through a causality and passivity checker before handing them to a time-domain or channel tool.
Question for the board: do you fit Djordjevic-Sarkar from a single datasheet point, or from your own measured test-coupon data?
1. Why constant values are wrong
Real dielectric permittivity and loss are linked by the Kramers-Kronig relations. If the loss is non-zero, the permittivity must change with frequency. A model with frequency-independent Dk and tan delta violates this, so time-domain results show pre-cursor ringing or a step response that starts before the signal arrives.
2. The Djordjevic-Sarkar (wideband Debye) model
It describes permittivity as a smooth function with a nearly constant loss tangent over a chosen frequency span (for example 1 kHz to 1 THz). You give it one Dk and tan delta at a reference frequency, and it returns a causal curve where Dk falls slowly with frequency and loss stays close to constant. Most commercial solvers offer it as a built-in option; openEMS and other open tools can approximate it with several Debye or Lorentz poles.
3. Picking the reference point
- Laminate datasheets often quote Dk at 1 or 10 GHz, sometimes from different test methods (stripline resonator, split post, clamped stripline). The values can differ by several percent.
- Use the value measured closest to your band, from the same test method if you have it, and note which one you used.
- Remember that the design Dk recommended by the vendor for impedance calculation may differ from the process Dk used for QC.
4. Glass weave and anisotropy
Woven glass laminates are anisotropic: in-plane Dk differs from the z-axis Dk. Microstrip and stripline fields mostly see z-axis Dk, but coupled lines and filters with strong horizontal fields can see the in-plane value. Spread-glass styles reduce the fibre weave skew in differential pairs.
5. Combining with roughness
Dielectric and conductor models must be fitted together. If you tune roughness to match measured loss while using a non-causal dielectric, you may get the right insertion loss for the wrong reason and the phase will still disagree.
6. Quick sanity check
Run the extracted S-parameters through a causality and passivity checker before handing them to a time-domain or channel tool.
Question for the board: do you fit Djordjevic-Sarkar from a single datasheet point, or from your own measured test-coupon data?