Facebook
Google
GitHub
Linkedin
differential line loss question
- Thread starter yef smith
- Start date
SDD21 is the differential-mode insertion loss in mixed-mode S-parameters — i.e. how much of a pure differential signal you get out at port 2 relative to what you put in at port 1. For an ideal lossless line it would be 0 dB and flat; the downward slope you see is the line's physical loss.
Three mechanisms produce that attenuation, and they scale differently with frequency: 1. Conductor (skin-effect) loss — at RF the current crowds into a thin surface layer, so effective resistance rises and loss grows roughly proportional to sqrt(f). This usually dominates the lower-to-mid part of the curve. 2. Dielectric loss — the laminate turns some of the field into heat; this grows roughly linearly with f and dominates at the high end. 3. Copper roughness — the etched surface increases the skin-effect loss beyond what a smooth-foil model predicts. That is why an SDD21 trace bends progressively downward rather than staying a straight horizontal line. In simulation you can fit it with the well-known causal model that separates a skin-effect term (proportional to sqrt(f)) and a dielectric term (proportional to f) — useful when you want to de-embed or extrapolate measured loss. One practical note: make sure you are reading SDD21 (the differential-to-differential response), not single-ended S21. It is derived from the four single-ended responses, e.g. SDD21 = 0.5 * (S21 - S43 - S23 + S41) for the standard port ordering — so any mode conversion (SCD21, from asymmetry) also steals energy from it. Keeping the two legs length- and impedance-matched is what keeps SDD21 close to the single-ended expectation and limits common-mode conversion. References worth reading: - Eric Bogatin, "Signal and Power Integrity – Simplified" (clear treatment of S-parameters and differential loss) - Howard Johnson & Martin Graham, "High-Speed Digital Design" - The VNA / mixed-mode S-parameter application notes from Keysight and Tektronix (free PDFs) for the SDD/SCD definitions and port ordering. If you tell me your stackup (dielectric, trace geometry and length) and the frequency range of the plot, I can help sanity-check whether the slope you're seeing is reasonable.
Three mechanisms produce that attenuation, and they scale differently with frequency: 1. Conductor (skin-effect) loss — at RF the current crowds into a thin surface layer, so effective resistance rises and loss grows roughly proportional to sqrt(f). This usually dominates the lower-to-mid part of the curve. 2. Dielectric loss — the laminate turns some of the field into heat; this grows roughly linearly with f and dominates at the high end. 3. Copper roughness — the etched surface increases the skin-effect loss beyond what a smooth-foil model predicts. That is why an SDD21 trace bends progressively downward rather than staying a straight horizontal line. In simulation you can fit it with the well-known causal model that separates a skin-effect term (proportional to sqrt(f)) and a dielectric term (proportional to f) — useful when you want to de-embed or extrapolate measured loss. One practical note: make sure you are reading SDD21 (the differential-to-differential response), not single-ended S21. It is derived from the four single-ended responses, e.g. SDD21 = 0.5 * (S21 - S43 - S23 + S41) for the standard port ordering — so any mode conversion (SCD21, from asymmetry) also steals energy from it. Keeping the two legs length- and impedance-matched is what keeps SDD21 close to the single-ended expectation and limits common-mode conversion. References worth reading: - Eric Bogatin, "Signal and Power Integrity – Simplified" (clear treatment of S-parameters and differential loss) - Howard Johnson & Martin Graham, "High-Speed Digital Design" - The VNA / mixed-mode S-parameter application notes from Keysight and Tektronix (free PDFs) for the SDD/SCD definitions and port ordering. If you tell me your stackup (dielectric, trace geometry and length) and the frequency range of the plot, I can help sanity-check whether the slope you're seeing is reasonable.
