My short version has always been: "tightly coupled" = as close to each other as realistically possible while keeping the characteristic impedance of the line. no breaks in the ground plane under the pair.
This should be good. My guesses: Myth 1 - differential pairs need to be routed close together, Myth 2 - return path "doesn't matter", Myth 3 - it doesn't matter where you match lengths so long as the full run is length matched, Myth 4 - you *need* to match lengths *exactly*, Myth 5 - there's something magical about 100 Ohms
@@Zachariah-Peterson from definition the differential pair do not need any reference plane but here we can see if we apply some reference plane then track width and space are easier to build. I hope Eric Bogatin explain this problems correctly.
For point 2, I hear so many different things on this topic. Rick Hartley says to treat Differential Pairs as 2 single ended lines, but you say otherwise. Which guru do I listen to? Thanks.
Rick Hartley and I are saying the exact same thing. What I am telling you is that in order to set one trace's impedance to a target value when there is no ground plane, you need to have the other trace close by because there is mutual inductance and mutual capacitance between the two traces. Once you do this, the signal on one trace will still be treated as its own signal. In the case where you have the ground plane, the odd-mode impedance of a single trace is what matters and what determines reflection from an individual pin as well as the propagation delay. The difference is that you do not need to have the 2nd trace close by just to get the impedance to a target value. The other differences between the two situations are radiation of common mode noise and creation/reception of differential crosstalk. EDIT: Rick Hartley sat in an audience of a PCB East presentation I gave earlier this year and affirmed everything I stated in my comment.
The Layer Stack Manager in Altium Designer uses the Simbeor 2D solver to calculate trace impedance and propagation constant without losses and dispesion. In frequency ranges where the dielectric constant is stable and losses are low, the results are very accurate the engine and runs in the backend of Altium Designer.
Engaging video! Most of these seem like "oversimplifications" rather than "myths" per se. One point of interest is in juxtaposing #2 "Differential pairs and ground" versus #5 "Differential pairs and length matching". In the latter, there's a subsidiary point about placing the length-tuning sections near to the inhomogeneity, so that we avoid "mode conversion". Mode conversion appears where the pair has been designed to achieve the target impedance through coupling of the two conductors (instead of proximity to ground plane) and the signals on the two wires get substantially out-of-step (one is delayed relative to the other). Now the two out-of-step signals just function more like two independent signals, with the coupling providing crosstalk between the two wires. Aka unwanted common-mode signal. OK, but that problem would not arise (or be far less) where the target impedance has been achieved primarily through proximity to ground plane (as described in #2), and not through interaction of the two conductors with each other. In that case, presumably one could place the tuning sections wherever one wanted with no ill effects. All that matters is that the total travel time along each path is a good match. I'm not saying that is something to try to do unnecessarily, just pointing out this interesting interaction between two of these "myths".
My short version has always been: "tightly coupled" = as close to each other as realistically possible while keeping the characteristic impedance of the line. no breaks in the ground plane under the pair.
This should be good. My guesses: Myth 1 - differential pairs need to be routed close together, Myth 2 - return path "doesn't matter", Myth 3 - it doesn't matter where you match lengths so long as the full run is length matched, Myth 4 - you *need* to match lengths *exactly*, Myth 5 - there's something magical about 100 Ohms
Actually myth 2 is "You always/never need ground"
@@Zachariah-Peterson so close!
@@Zachariah-Peterson from definition the differential pair do not need any reference plane but here we can see if we apply some reference plane then track width and space are easier to build. I hope Eric Bogatin explain this problems correctly.
Informative as always
For point 2, I hear so many different things on this topic. Rick Hartley says to treat Differential Pairs as 2 single ended lines, but you say otherwise. Which guru do I listen to? Thanks.
Rick Hartley and I are saying the exact same thing. What I am telling you is that in order to set one trace's impedance to a target value when there is no ground plane, you need to have the other trace close by because there is mutual inductance and mutual capacitance between the two traces. Once you do this, the signal on one trace will still be treated as its own signal. In the case where you have the ground plane, the odd-mode impedance of a single trace is what matters and what determines reflection from an individual pin as well as the propagation delay. The difference is that you do not need to have the 2nd trace close by just to get the impedance to a target value. The other differences between the two situations are radiation of common mode noise and creation/reception of differential crosstalk.
EDIT: Rick Hartley sat in an audience of a PCB East presentation I gave earlier this year and affirmed everything I stated in my comment.
Is simbior included in an Altium product?
The Layer Stack Manager in Altium Designer uses the Simbeor 2D solver to calculate trace impedance and propagation constant without losses and dispesion. In frequency ranges where the dielectric constant is stable and losses are low, the results are very accurate the engine and runs in the backend of Altium Designer.
Engaging video! Most of these seem like "oversimplifications" rather than "myths" per se. One point of interest is in juxtaposing #2 "Differential pairs and ground" versus #5 "Differential pairs and length matching". In the latter, there's a subsidiary point about placing the length-tuning sections near to the inhomogeneity, so that we avoid "mode conversion". Mode conversion appears where the pair has been designed to achieve the target impedance through coupling of the two conductors (instead of proximity to ground plane) and the signals on the two wires get substantially out-of-step (one is delayed relative to the other). Now the two out-of-step signals just function more like two independent signals, with the coupling providing crosstalk between the two wires. Aka unwanted common-mode signal.
OK, but that problem would not arise (or be far less) where the target impedance has been achieved primarily through proximity to ground plane (as described in #2), and not through interaction of the two conductors with each other. In that case, presumably one could place the tuning sections wherever one wanted with no ill effects. All that matters is that the total travel time along each path is a good match. I'm not saying that is something to try to do unnecessarily, just pointing out this interesting interaction between two of these "myths".
Can you make video, how to make library componen