Post by w0qe on Dec 7, 2009 22:16:46 GMT -5
Recently I ran across an issue with the AIM4170 and the VNA2180 as a beta tester which can cause the measured data to be incorrect. Bob, W5BIG, is aware of the issue and is working on software to minimize its occurrence. The issue affects every impedance analyzer product and has always existed.
The issue:
All the W5BIG products have a very nice feature that allows the data obtained during calibration at the predetermined cal. frequencies to be interpolated to other frequencies when doing scans. For example a calibration starting at 1MHz, stopping at 180MHz, with steps every 1MHz will save data at 1, 2, 3, 4, ....179, and 180MHz. Then if you can do a scan starting at 5MHz, stopping at 10MHz, with steps every 500KHz, data will be collected at 5, 5.5, 6, 6.5, 7....9.5, and 10MHz. Some of these frequencies have cal. data available but others such as 5.5MHz use the cal. data from 5MHz and 6MHz via an averaging algorithm called interpolation.
The situation becomes more complicated when there is an external circuit in the calibration path. This can be a piece of coax or a filter. In this case, the impedances seen at the analyzer can change rapidly from sample to sample during calibration and are a function of external cable length, frequency, circuitry in the cal path, step sizes etc. such that the interpolation can't be guaranteed to give correct results all the time. I own an HP VNA that was state of the art in the 1990s and it does NOT allow interpolation without putting a warning on the screen indicating that the data may be inaccurate. I had never previously seen the issue since my big VNA had "trained me" to almost always scan with the same data points as I had used for calibration. Most users don't even pay attention to whether the data is being interpolated or not and the analyzer works great most all the time. However, characterizing in advance whether the data will be correct or not is extremely hard to do.
Work around for the AIM and the VNA:
(these techniques can be beneficial for all equipment of this type)
1.) Always scan using exactly the same frequencies as were used during calibration. This is the best solution and gives the most accurate data. However, it is quite restrictive. Note that you don't need to use all the cal frequencies in the scan but that all scan frequencies need to also be a cal frequency.
2.) Increase the number of cal points (reduce the step size). Interpolation is much more accurate when the cal frequencies are spaced closer together. Use custom calibration which uses a more advanced algorithm for the interpolation.
3.) After loading a cal file and setting the desired scan limits, do a scan on the load resistor that was used for calibration. The resulting data should be a flat resistance of the correct value (or very close). This test appears to detect if the issue is going to occur but I can't be positive that this test is 100% perfect.
Solution:
Bob is currently working on having the software help the user by encouraging the user to scan using frequencies that were also cal frequencies and when interpolation is active, an indication will be present on the screen. Possible data interpolation issues will be flagged during calibration. The goal is to encourage the use of more accurate data but still allow for operation similar to what is done today.
To reproduce the issue:
1.) Attach a 36" piece of coax to the analyzer (other lengths have the same effect and may create aberrant data at different frequencies). In general the aberrations are related to the quarter wave points.
2.) Either do a standard cal. or a custom cal. with Fstart = 1MHz, Fstop = 180MHz, Fstep = 1MHz. and save the cal. file. The standard calibration routine always uses 1MHz increments. For custom cal the step size is optional but for this comparison study, use 1MHz so the results from the two calibration techniques can be compared side by side.
3.) Using the new cal file, set the scan limits to Fstart=1MHz, Fstop = 180MHz, Fstep = 1MHz and scan the resistor used for calibration through the same piece of coax used for the calibration. Turn on appropriate plot parameters and notice that the data looks good.
4.) Now go into the scan limits window and change Fstep to 0.5MHz and rescan the resistor. You will see incorrect data at 44 to 52MHz and at harmonics of these frequencies. Closer examination reveals that every other data point in the range where the issue occurs is good and that only scan frequencies that used interpolated data are incorrect. The glitches will occur when using either standard cal or custom cal although the custom cal will probably have smaller glitches because it uses a different interpolation algorithm. A screen shot showing the actual results is at www.w5big.com/DataIssueScan.gif
I passed this by Bob before posting it and he added the following comments:
The first major glitch occurs near the quarter wave frequency and repeats at subsequent half wave intervals. There is a lesser glitch at the first half wave point. This second glitch is more noticeable in the Return Loss trace because the return loss function is very sensitive to aberrations in the raw data. If you are only interested in frequencies well away from these critical points, the data is ok.
Interpolation has been used for hundreds of years and there are many different techniques. The main assumption is that data values do not change very much from one point to the next and an in-between point can be determined by using the known data points near it.
This assumption of slowly changing data is good for many useful cases where the AIM and the VNA are applied. If you calibrate with the standard open, short, resistor loads attached directly to the RF connector, you will see the data in the cal file doesn't change much from one step to the next. In this case interpolation works very well.
This phenomenon does not affect the measurement of the impedance at the input end of a transmission line to an antenna. However, it does apply when the calibration loads are placed at the far end of the transmission line for the purpose of canceling out the line and measuring impedance directly at the feed point of the antenna. In actual practice, the line does not have to be very long, even three feet can affect measurements in the VHF range.
Whenever it is practical, use the original cal data points for the scan. Some commercial instruments restrict their scans to only the cal data points. (This is accurate but it can be a nuisance when the application doesn't have a problem to begin with. It requires recalibrating when the setup is changed.)
The original cal data parameters used for custom cal can be reviewed up by using the Help->Status menu. The start and end points don't have to be the same as the original cal parameters. For example, if you cal from 3MHz to 8MHz with 0.01MHz steps, then you can change the start freq to 6.9MHz and step by 0.01MHz and you will still be using the original cal data to achieve the best results. The end freq doesn't matter.
When you do need to use in-between points, compare the results with a test scan using the coarser calibration step size. If aberrations are present they are not small errors, they are noticeable glitches, so you can tell if there is a potential problem. If the trace looks smooth with different step sizes, then interpolation is not an issue in this test range.
73,
Larry, W0QE
The issue:
All the W5BIG products have a very nice feature that allows the data obtained during calibration at the predetermined cal. frequencies to be interpolated to other frequencies when doing scans. For example a calibration starting at 1MHz, stopping at 180MHz, with steps every 1MHz will save data at 1, 2, 3, 4, ....179, and 180MHz. Then if you can do a scan starting at 5MHz, stopping at 10MHz, with steps every 500KHz, data will be collected at 5, 5.5, 6, 6.5, 7....9.5, and 10MHz. Some of these frequencies have cal. data available but others such as 5.5MHz use the cal. data from 5MHz and 6MHz via an averaging algorithm called interpolation.
The situation becomes more complicated when there is an external circuit in the calibration path. This can be a piece of coax or a filter. In this case, the impedances seen at the analyzer can change rapidly from sample to sample during calibration and are a function of external cable length, frequency, circuitry in the cal path, step sizes etc. such that the interpolation can't be guaranteed to give correct results all the time. I own an HP VNA that was state of the art in the 1990s and it does NOT allow interpolation without putting a warning on the screen indicating that the data may be inaccurate. I had never previously seen the issue since my big VNA had "trained me" to almost always scan with the same data points as I had used for calibration. Most users don't even pay attention to whether the data is being interpolated or not and the analyzer works great most all the time. However, characterizing in advance whether the data will be correct or not is extremely hard to do.
Work around for the AIM and the VNA:
(these techniques can be beneficial for all equipment of this type)
1.) Always scan using exactly the same frequencies as were used during calibration. This is the best solution and gives the most accurate data. However, it is quite restrictive. Note that you don't need to use all the cal frequencies in the scan but that all scan frequencies need to also be a cal frequency.
2.) Increase the number of cal points (reduce the step size). Interpolation is much more accurate when the cal frequencies are spaced closer together. Use custom calibration which uses a more advanced algorithm for the interpolation.
3.) After loading a cal file and setting the desired scan limits, do a scan on the load resistor that was used for calibration. The resulting data should be a flat resistance of the correct value (or very close). This test appears to detect if the issue is going to occur but I can't be positive that this test is 100% perfect.
Solution:
Bob is currently working on having the software help the user by encouraging the user to scan using frequencies that were also cal frequencies and when interpolation is active, an indication will be present on the screen. Possible data interpolation issues will be flagged during calibration. The goal is to encourage the use of more accurate data but still allow for operation similar to what is done today.
To reproduce the issue:
1.) Attach a 36" piece of coax to the analyzer (other lengths have the same effect and may create aberrant data at different frequencies). In general the aberrations are related to the quarter wave points.
2.) Either do a standard cal. or a custom cal. with Fstart = 1MHz, Fstop = 180MHz, Fstep = 1MHz. and save the cal. file. The standard calibration routine always uses 1MHz increments. For custom cal the step size is optional but for this comparison study, use 1MHz so the results from the two calibration techniques can be compared side by side.
3.) Using the new cal file, set the scan limits to Fstart=1MHz, Fstop = 180MHz, Fstep = 1MHz and scan the resistor used for calibration through the same piece of coax used for the calibration. Turn on appropriate plot parameters and notice that the data looks good.
4.) Now go into the scan limits window and change Fstep to 0.5MHz and rescan the resistor. You will see incorrect data at 44 to 52MHz and at harmonics of these frequencies. Closer examination reveals that every other data point in the range where the issue occurs is good and that only scan frequencies that used interpolated data are incorrect. The glitches will occur when using either standard cal or custom cal although the custom cal will probably have smaller glitches because it uses a different interpolation algorithm. A screen shot showing the actual results is at www.w5big.com/DataIssueScan.gif
I passed this by Bob before posting it and he added the following comments:
The first major glitch occurs near the quarter wave frequency and repeats at subsequent half wave intervals. There is a lesser glitch at the first half wave point. This second glitch is more noticeable in the Return Loss trace because the return loss function is very sensitive to aberrations in the raw data. If you are only interested in frequencies well away from these critical points, the data is ok.
Interpolation has been used for hundreds of years and there are many different techniques. The main assumption is that data values do not change very much from one point to the next and an in-between point can be determined by using the known data points near it.
This assumption of slowly changing data is good for many useful cases where the AIM and the VNA are applied. If you calibrate with the standard open, short, resistor loads attached directly to the RF connector, you will see the data in the cal file doesn't change much from one step to the next. In this case interpolation works very well.
This phenomenon does not affect the measurement of the impedance at the input end of a transmission line to an antenna. However, it does apply when the calibration loads are placed at the far end of the transmission line for the purpose of canceling out the line and measuring impedance directly at the feed point of the antenna. In actual practice, the line does not have to be very long, even three feet can affect measurements in the VHF range.
Whenever it is practical, use the original cal data points for the scan. Some commercial instruments restrict their scans to only the cal data points. (This is accurate but it can be a nuisance when the application doesn't have a problem to begin with. It requires recalibrating when the setup is changed.)
The original cal data parameters used for custom cal can be reviewed up by using the Help->Status menu. The start and end points don't have to be the same as the original cal parameters. For example, if you cal from 3MHz to 8MHz with 0.01MHz steps, then you can change the start freq to 6.9MHz and step by 0.01MHz and you will still be using the original cal data to achieve the best results. The end freq doesn't matter.
When you do need to use in-between points, compare the results with a test scan using the coarser calibration step size. If aberrations are present they are not small errors, they are noticeable glitches, so you can tell if there is a potential problem. If the trace looks smooth with different step sizes, then interpolation is not an issue in this test range.
73,
Larry, W0QE