It shouldn't matter if the port transmission lines have 50 Ohm characteristic impedance (assuming your Wilkinson is designed to match 50 Ohms). Have you checked the port mode solution? You can do so after running the solver by clicking on "2D/3D Results --> Port Modes --> Port 1 --> e1" (for port 1) and looking at the information window that pops up where it says "Line Imp. [Ohms]".

If that number is substantially different from 50 Ohms, then by changing the port length you may be inadvertently introducing an impedance transformation (thus spoiling your return loss). Assuming this is all in microstrip, and the line impedance is too low, then you need to make your line narrower. If it is too high, make it wider.

Matt

Thanks Mattew. All your assumption are correct, the output ports are microstrips designed to be 50ohms. I checked the input impedance is 49.48ohm and the output impedance is 49.27 ohms. I do not think this is the problem. What do you think ? 

Hmm, that's strange. When you say the return loss is 2dB less when you extend the output ports, do you mean 2dB better? And are you looking at the return loss at the output port? If so, then I suppose it is possible the extra line length has 1 dB of loss, and that would improve the output return loss by 2 dB, almost as if you had a 1 dB pad on each port.

But that's probably reaching. Only other thing I can think of off-hand is that the ports were too close to the circuit to begin with and you were truncating the field solution around the active sections. You want to make sure that the ports are far enough away that all the evanescent fields can decay to a negligible level at the outer boundary. I usually get at least a quarter- to a half-wavelength away. Then you can use the reference plane extension on the ports to de-embed back to the inner part of the structure.

Are the output ports on opposite sides of the boundary, or are they parallel running to the same boundary? If the latter, there could be an issue with them coupling to each other.

Sorry if I am telling you things you already know. Just grasping at some straws. If you wanted to post a screen capture of your model (before and after doubling the outer port length) I would be happy to take a look and see if anything stands out.

Best,

Matt

Thanks Mathew.. you highlighted important point which I checked. I attached the two design before and after extending the output ports. What I meant by  2dB worse return loss, is that it was  -18dB and become -15.7dB. Beside I am measuring the return loss from the input side. The distance between the last stage and the output port part which is going down is about half wave length. You can see from the design that the output ports are in opposite directions.

Could you please explain in more details what you meant by "  Then you can use the reference plane extension on the ports to de-embed back to the inner part of the structure."

Thanks a lot

Hi Ammar,

Sorry, I don't see your designs, did you forget to attach them?

About the reference plane extension: If you open the port properties (the same window that comes up when you first define the port), about mid-way down there is a field labelled "Distance to ref. plane." Enter a negative number here equal to the length of the output transmission line. You should see some extension lines appear on the model.

CST will then solve the model as normal, but prior to displaying the s-parameters, it will roll back the phase from the port s-parameters (and divide out any small amount of loss) assuming a uniform transmission line equal to the cross section it sees at the actual port. This makes it look as if the port is right next to the power divider, but without truncating the fields around it by the walls of the bounding box.

CST likes this number to be negative because logically you are subtracting phase from the EM solution rather than adding it. If you put a positive number here, it is equivalent to adding some extra transmission line to the circuit without actually increasing the size of the model it has to solve. I've never had reason to do that myself, I almost always use negative numbers here.

I hope that makes sense.

Matt

Sorry I forgot to upload them. 

Thanks! Ok, now that I see it, I understand that you not only made the port lines longer, but you folded them as well. I thought they were just longer, but straight. This is a whole different story.

That being the case, I would wonder if the degradation in return loss isn't due to the bends, or possibly coupling between the main power divider and the parallel sections of the output lines (the parts running vertically in your attached images). It wouldn't take much to get a result of -15.7 dB. I see you have the corners mitered, but are they optimally mitered? Even if they were, I wouldn't be 100% confident that is frequency-independent, especially if the substrate is relatively thick for your frequency. You could check by running a simulation of just the output lines connected together in an upside-down "U" without the Wilkinson, as shown in the attached image. If this result is not better than, say, 25 dB, it could be your problem. (Any less than that, and the standing wave created between the bends and the 18 dB return loss of the Wilkinson could add constructively and give you the ~16 dB return loss at particular frequency points. A good rule of thumb to remember is that two discontinuities of equal magnitude combine such that the net return loss is 6 dB worse. So if your mitered lines alone gave you 18 dB, and your Wilkinson is also 18 dB, then jointly it would be only 12 dB!)

Another possibility is that the horizontal parts of the output lines are fairly close to the edge of the substrate. That in itself didn't change between your two models, but it does look like the north wall of the bounding box moved -- because in the first image the width of the ports pushed it out, whereas in the second image the ports are closer to the center of the substrate, so it appears the bounding wall moved in to the edge of the substrate. As we discussed before, its important to have the bounding box far enough away that the all the fields around the active parts of your structure are not significantly truncated.

I hope this helps.

Matt

Thanks to you. 

Yes the corners optimally mitered. I am using a rogers 3035 substrate which thickness is 0.75mm ( Is this considered thick or thin relatively highest frequency the power divider working on which is 10.6 GHz ? )

I simulated the output ports and got a return loss below 25dB from 1-7.7GHz and reached rapidly as you can see from attached image ( the red line). Also I increased the upper boundary edge of the boundary box ( and got the blue line). It seems the upper boundary box effecting the results. 

I also increased the upper boundary box for the second Wilkinson design and got an enhancement as you can see from the second image( the red line after increasing the boundary box size). It seems to me that the effect of boundary box size is not similar on the Wilkinson second design and the the U shape output design. It enhanced the return loss but did not reach the firs Wilkinson design return loss ( the pink line in the second attachment) might be the rule of thumb you mentioned ? 

Hi Ammar,

I think your thickness is ok.

I wouldn't have expected the return loss of the U or its improvement upon moving the boundary to correlate directly with a similar improvement in the input return loss of the WIlkinson, it was just a approximate test to see how well your miters were working. When looking at s11 of the WIlkinson, the two legs of the U are excited in phase with each other, which is a little different when propagating a wave from side to side.

In any case, I think you are seeing that the boundary box is having an effect. I suspect you may also have some coupling between the vertical sections of the output legs and the power divider. You could try moving those out a line width and seeing if that has an effect.

Matt

Instead of mitred corners in your extended version, have you tried smooth curves? That  should be easy enough in a sim. The mitres look a little on the fat side from the illustrations, you could try removing more material; the current crowding on the inside of the corner means that the surplus material is a little capacitive. This is a time when an intuitive half diagonal mitre doesn't work well - try a 70% cut.  I'm assuming that you have built this and are not just working with a sim solution.

It should be noticed that "-15dB" or "-17dB" is a very small value. in this case, although the return loss of the corner has been optimized to -25 dB, the returned EM energy still may be enough to cause the increase of the return loss of the whole power devider. I suggest you to calculate it carefully. 

In addition, there is"fixed 50ohm normalization"  item in the CST simulation. Do you choose it? Choosing it may cause additional simulated return loss.

Thanks for all... I optimized the structure and got -20 dB return loss over the band ..... know I will work on my next step which is combining three of 1 to 2 power divider to get 1 to 4. 

Thanks Matthew for the detailed information.