TRL校准精度高，不匹配

从理论上讲，仅基于校准标准的质量，应该去除不匹配的1端口测量，我们可能会对剩余的不匹配进行校准。
但对于2端口测量，直通测量的准确性取决于端口RAW不匹配和质量校正残差匹配的乘积，因此任一端口的不匹配都会降低2端口（S21）测量值，
相同质量的校准套件。
此外，即使在一个端口测量中，如果测试端口和固定装置之间存在任何不稳定性（例如电缆），则不稳定性将导致校正的性能降级得更快，因为不良匹配比良好的匹配夹具更快。



     以下为原文

  Theoretically, the mismatch should be removed for 1 port measurements based only on the quality of the calibration standards, which we might cal the residual mismatch.  But for 2 port measuerments, the accuracy of the the through measurement depends upon the product of the RAW mismatch at the port and the quality corrected residual match, so a bad match at either port will degrade your 2-port (S21) measurements, for the same quality of calibration kit.

Further, even in one port measuerments, if you have any instability (such as a cable) between the test port and the fixture, the instabilty will cause the corrected to performance to degrade much more quickly for a bad match than a good match fixture.

谢谢你的回复。
我们可以量化退化吗？
假设我的标准在50 GHz时仅有5 dB的回波损耗。
是否有特定的测试单元可以放在晶圆上以帮助确定校准质量？
再次感谢您的帮助。



     以下为原文

  Thanks for the response. Can we quantify the degradation? Say if my standards only have 5 dB of
return loss at 50 GHz. Is there a particular test cell that could be put on a wafer to help determine
the quality of the calibration?

Thanks again for the help.

这真的很难，因为这一切都归结为线路Z0的质量。
对于简单的情况，如果用作参考线的线Z0通过某个阻抗（例如5欧姆）出错，那么剩余源和负载匹配的误差相同。
Zline-Z0 /（Zl + Z0）给出线路回波损耗的公式，5欧姆给出约0.05的消除误差或-26 dB，称之为Gamma_Error，或者Ge的S21误差为2 *（SMraw * Ge + LMraw * Ge
）其中SMraw是端口1原始匹配，LMraw是端口2原始匹配。
校准后，通过查看源匹配和加载匹配错误术语，可以非常好地估计原始匹配。
在您的情况下，对于源和负载上的5 dB原始匹配，得出S21_err = 4 *（0.562 * .05）或以dB为单位20log10（S21_err + 1）= 0.925 dB;
对不起的结果感到抱歉。
下次更好的比赛。
注意，如果你有10 dB原始匹配和40 dB线（1欧姆误差），你的S21误差将是0.11 dB！



     以下为原文

  It's really difficult because it all comes down to the quailty of the Z0 of the line.  For the simple case, if the line Z0 used as the reference line is in error by some impedance (say 5 ohms) then the residual source and load match are in the same error.  Zline-Z0/(Zl+Z0) gives formula for line return loss, and 5 ohms gives about 0.05 relfection error or -26 dB, call it Gamma_Error, or Ge

the S21 error is then 2*(SMraw*Ge+LMraw*Ge) where SMraw is port 1 raw match and LMraw is port 2 raw match.  After the calibration, you get a very good estimate of the raw match by looking at the source match and load match error terms.

In your case, for 5 dB raw match on both source and load, that gives S21_err=4*(0.562*.05) or in dB 20log10(S21_err+1)=0.925 dB; sorry for the bad result.  Better match next time.

Note, if you have a 10 dB raw match, and a 40 dB line (1 ohm error) you S21 error would be 0.11 dB!

这正是我一直在寻找的定量答案。
S11和S22错误是否有类似的等式？
此外，从等式中可以看出，只要我有一个完美的50欧姆线用于标准，就不会出现一组可怕的电缆或固定装置的错误。
我认为情况并非如此。
再次感谢你的帮助。



     以下为原文

  That's exactly the quantitative answer I was looking for. Is there a similar equation for the S11 and S22 errors? Also,
it seems from the equation, that as long as I have a perfect 50 ohm line for the standard, there would be no error
from a horribly matched set of cables or fixtures. I assume this not to be the case.

Thanks again for all your help.

不，这是特别的情况。
如果标准是完美的，并且没有噪声或不稳定性，则残留误差将为零，即使系统的匹配很糟糕，校准也将是完美的。
Theorectically。
但是，如果系统有任何不稳定性或噪声（电缆，本底噪声），那么测量结果将无法完全重复，然后最小的可执行残余误差将受到本底噪声和可重复性的限制。
事实上，最好的TRL校准就是这种情况;
连接器的可重复性不如标准（TRL线路阻抗的不确定性非常小），因此残余误差仅限于连接器的可重复性，没有0.0005线性（约-65 dB）的数量级。
如果我再多花一点时间，我会发布S11和S22的错误，或者你可以查看我的书，第190页



     以下为原文

  No, that is exacly the case.  If the standards are perfect, and absent noise or instability, the residual errors will be zero and the calibration will be perfect even if the match of the system is terrible.  Theorectically.

But, if the the system has any instability or noise (cables, noise floor) then of coursed the measuerments will not be perfectly repeatable and then the minimum acheivable residual errors will be limited by noise floor and repeatablity.  In fact that is the case with the best TRL calibrations; the connector repeatabilty is not as good as the standards (the uncertainy of the TRL line impedance is very small) and so the residual error is limited to the connecctor repeatability, no the order of 0.0005 linear (about -65 dB).

If I get a little more time, I'll post the error for S11 and S22, or you could just look in my book, page 190