开云体育

Whether you do the calibration on a single port or on both, there are errors in the .cal file. To put it simply, we will say that these are parasites that generate false measurements. The real cause is unknown to me and it doesn't really matter. Either way, there will be mistakes.

The .cal file can be corrected by hand. I enter this file into Excel and, by arithmetic operations, I can locate errors. In the same column, a value is not independent of the preceding and the following; it is therefore quite easy to detect and locate abnormal values. It is then possible to replace the erratic value by an interpolation of the value before and the value after. So far I have never encountered two consecutive false errors in the same column. It can be assumed that, in this case, the algorithm proposes to carry out a new calibration. By hand, we are not sure to have identified all the faults. It's tedious and unreliable.

It would be necessary to develop an algorithm performing this detection and replacement. With Excel, I make line-to-line comparisons, for example by comparing the difference in value between two lines with a threshold. When a value (in a column) is false on line N, it causes a doubt on lines N-1 and N+1.

The notion of fixed threshold is never satisfactory in a good algorithm; this threshold, if there is a threshold, must be adjusted dynamically. The values ??follow pseudo sinusoidal curves since it is the frequency which varies from line to line; this property must be exploited.

The problem of processing the first and last lines of the file arises.

We perceive that there is, either in the box, or in nanovna-saver, a filtering based, at least in part on an average. Measurement faults are not marred by noise; they are very false (erratic). It follows that taking into account an average value, somewhere, to eliminate measures does not work well.

One of the difficulties is that it is a question of correcting the .cal file which is the result of a processing parameterized by

Number of measurements to average
Number to discard

I don't even see which parameters would be the most effective to facilitate post processing.

73
--
F1AMM (Fran?ois)