|
Since most filters are a series of resonators of some kind or another, terminating them in a resistance other than the design resistance will probably change the filter characteristics. Consider a filter that effectively has an input that is a RLC circuit, where the R is the terminating impedance of the source. If you change R from, say, 300 ohms to 50 ohms, then the Q will be different. That will certainly change the skirts, and will also probably change the overall passband (since most filters are stacked up responses of multiple resonances).
|
Hi
it seems to me that for a linear or quasi-linear circuit without active elements, it should be correctly compensated by single precision floating point calculation using the Z renormalization option and this with the simplest possible hardware tricks during the measurements, thus avoiding hardware imperfections of the ferrite core or resistance transformers and without any calibration alteration.
May be Jhon can illustrate for us response comparaison between H4+Z renormalization and oxilloscope classic measurements of the same ceramic filter. it will be very appreciated.
73's Nizar .
|
Jim,
Of course the change in termination impedance changes the filter response. The point of the renormalization is that the response can be recalculated to show what it would be at the different impedance. Doing so requires knowledge of the transfer function and the impedance of both ports.
Similar case: If you know the open circuit voltage and output impedance of a source, you can compute its output level into any impedance.
--John
toggle quoted message
Show quoted text
On Fri, Feb 14, 2025 at 09:42 AM, Jim Lux wrote:
Since most filters are a series of resonators of some kind or another,
terminating them in a resistance other than the design resistance will
probably change the filter characteristics. Consider a filter that effectively
has an input that is a RLC circuit, where the R is the terminating impedance
of the source. If you change R from, say, 300 ohms to 50 ohms, then the Q
will be different. That will certainly change the skirts, and will also
probably change the overall passband (since most filters are stacked up
responses of multiple resonances).
|
Nizar,
I put series 390 ohm resistors on both port 0 and port 1, creating a 440 ohm environment for the filter, close to the 430 ohm renormalization I showed before (see message 39460).
I then normalized the S21 measurement by connecting the two resistors end for a thru measurement.
Attached is the filter S21 magnitude in the 440 ohm environment. It matches very well with the Z=430 renormalized measurement I posted before, allowing for the loss in dynamic range due to the resistors.
--John
toggle quoted message
Show quoted text
On Fri, Feb 14, 2025 at 10:28 AM, Team-SIM SIM-Mode wrote:
Hi
it seems to me that for a linear or quasi-linear circuit without active
elements, it should be correctly compensated by single precision floating
point calculation using the Z renormalization option and this with the
simplest possible hardware tricks during the measurements, thus avoiding
hardware imperfections of the ferrite core or resistance transformers and
without any calibration alteration.
May be Jhon can illustrate for us response comparaison between H4+Z
renormalization and oxilloscope classic measurements of the same ceramic
filter. it will be very appreciated.
73's Nizar .
|
But renormalization just changes the calculation for the S parameters.
It doesn't fix the change in Q.
toggle quoted message
Show quoted text
-----Original Message----- From: < [email protected]> Sent: Feb 14, 2025 4:20 PM To: < [email protected]> Subject: Re: testing non-50 ohm filters was Re: [nanovna-users] NanoVNA port renormalization Jim, Of course the change in termination impedance changes the filter response. The point of the renormalization is that the response can be recalculated to show what it would be at the different impedance. Doing so requires knowledge of the transfer function and the impedance of both ports. Similar case: If you know the open circuit voltage and output impedance of a source, you can compute its output level into any impedance. --John On Fri, Feb 14, 2025 at 09:42 AM, Jim Lux wrote:
Since most filters are a series of resonators of some kind or another,
terminating them in a resistance other than the design resistance will
probably change the filter characteristics. Consider a filter that effectively
has an input that is a RLC circuit, where the R is the terminating impedance
of the source. If you change R from, say, 300 ohms to 50 ohms, then the Q
will be different. That will certainly change the skirts, and will also
probably change the overall passband (since most filters are stacked up
responses of multiple resonances).
|
Jim,
It renormalization doesn't change the Q in the measured test circuit, but it can compute what the Q would be at the new impedance.
The S parameters are a complete description of a linear 2-port, allowing (with enough calculation) the prediction of the 2-port behavior in any impedance environment.
--John
toggle quoted message
Show quoted text
On Fri, Feb 14, 2025 at 05:31 PM, Jim Lux wrote:
But renormalization just changes the calculation for the S parameters.
It doesn't fix the change in Q.
-----Original Message-----
From: <[email protected]>
Sent: Feb 14, 2025 4:20 PM
To: <[email protected]>
Subject: Re: testing non-50 ohm filters was Re: [nanovna-users] NanoVNA port
renormalization
Jim,
Of course the change in termination impedance changes the filter response. The
point of the renormalization is that the response can be recalculated to show
what it would be at the different impedance. Doing so requires knowledge of
the transfer function and the impedance of both ports.
Similar case: If you know the open circuit voltage and output impedance of a
source, you can compute its output level into any impedance.
--John
On Fri, Feb 14, 2025 at 09:42 AM, Jim Lux wrote:
Since most filters are a series of resonators of some kind or another,
terminating them in a resistance other than the design resistance will
probably change the filter characteristics. Consider a filter that
effectively
has an input that is a RLC circuit, where the R is the terminating impedance
of the source. If you change R from, say, 300 ohms to 50 ohms, then the Q
will be different. That will certainly change the skirts, and will also
probably change the overall passband (since most filters are stacked up
responses of multiple resonances).
|
Hi John,
Thanks for running this comparison test. Looking at all 3 of your plots the renormalization routine did an amazing job compared with no renormalization in which no series resistors were used. While there are some slight differences between your renormalization plot and your series resistor plot they are indeed very similar and not sure how much of the differences are attributable to differences in dynamic range or fixturing, etc. (but all in all they compare much better than I expected).
Differences noted between renormalization and the series resistance method as follows:
Looking at the bandwidth around the 20 dB bandwidth area there is a slight difference (bandwidth slightly greater using the series resistors).
The left skirt when using renormalization has a more noticeable inward kink in it.
The insertion loss looks worse when using the series resistor method (not much but noticeable).
Bottom line is that while the renormalization and series resistor method did not provide the exact same results, they are indeed very close (really surprising in my opinion).
P.S. I wonder how well the series resistor method compares with the L pad matching pad or transformer matching method but I can run that test on my own.
Thanks for running the test.
Don
|
I've observed that reversing just about any Murata 10.7 MHz ceramic filter alters the response and changes the measured distortion in an FM tuner. I believe this means that S11 differs from S22, or S21 differs from S12, or both. Evidently this precludes using a NanoVNA to accurately renormalize from 50 to 330 ohms since it only measures S11 and S21. But if you reverse the filter in the measurement circuit, you can measure S22 and S12. Then if you combine all four measurements, you can use the full renormalization equations for S11, S21, S12, and S22 to get the correct response.
Since the NanoVNA won't do this itself, I've written a little Windows program to do it. You feed it two 50 ohm .s2p files and it generates an .s2p file for whatever renormalization Z you specify. It seems to work. When I get it all documented, I'll post a link.
Brian
|
The renormalization program is listed at the top of this page:
See the bottom of the page for downloading instructions.
I think the program works, but I don't have a NanoVNA to fully test it. I'm counting on some intrepid soul to make a forward and reverse filter measurement and let me know how it goes. I don't think this is the place to debug software. Contact me at the email address at the bottom of the page given above.
I tried an example from the source of the renormalization equations, but results were a little off. I had to manually invert a matrix to determine that the results given in the source were wrong!
Brian
|
Hi Brian
May be Jhon can reverse the same ceramic filter and get a comparative response with forward curve already succesfully published here , i expect to found almost the same response, indeed the ceramic filter is loaded with same virtual impedance of 430 Ohm on both sides.
Thanks John for sharing your superbe Z-renormalisation experiments.
73's Nizar .
|
Hi Brian and Nizar,
As a reference I went and tested my 10.7 MHz ceramic filter connected in both directions and see very little difference (part number SFE10.7MA5-A which is a Murata part number). See attached plot showing both directions tested (the plots are overlayed, and they really fall right on top of each other).
Also note that this ceramic filter input and output impedance is advertised as 330 ohms but it looks like my transformers are presenting an impedance of approximately 280 ohms to the ceramic filter input and output based on measurements I did with my NanoVNA-F when the transformers were connected to a 50 ohm load (should have been 313 ohms based on my transformer turns ratio but my measurements yielded 280 ohms).
Note: the Red marker in my attached plot is at 10.700 MHz and this was indeed the frequency where the response curves were peak.
Just FYI,
Don
|
Thanks for the test, Don. I have dozens of Murata 10.7 MHz filters from 110 to 280 kHz bandwidth. Some show little difference when reversed, but most do. Incidentally, it is probably the difference in group delay that causes the difference in audio distortion I observe when reversing a filter. FM detectors are sensitive to group delay. Try more filters if you can. If you could post forward and reverse .s2p files for 50 ohm drive, that would help me check my program.
|
On Sun, Feb 16, 2025 at 07:46 AM, Donald Kirk wrote:
SFE10.7MA5-A
Don, attached are the specs I have for your filter. Note that the G.D.T. entry is blank. GDT means group delay time and this filter has no spec. Murata made many special types for FM IFs where group delay variation over the passband is specified. These filter types reduce detected audio distortion. The body of these filters is usually blue. I've also attached the Murata test circuit. Note that the capacitor is specified on just one side of the filter. I assume its purpose is to account for the input capacitance of the stage the filter drives. Note that Murata specifies a tolerance of 2 pF for the 10 pF. I've attached curves that show response variation for various capacitive loads for a 230 kHz filter. Brian
|
Nizar,
With only S11 and S21 measured, the renormalization depends on the symmetry of the device measured.
Most passive linear devices have S21=S12. (Isolators are an exception.)
Many filters have S11 about equal to S22, and the renormalization will succeed based on just how close to equal they are.
Filters with significantly different input and output impedances will not be properly renormalized without measurement of all 4 S-parameters.
Some ceramic filters do show noticeable S11/S22 differences when reversed, while others look pretty close.
I wouldn't depend on renormalization on the NanoVNA for something like device acceptance where good accuracy is needed.
--John
toggle quoted message
Show quoted text
On Sat, Feb 15, 2025 at 11:44 PM, Team-SIM SIM-Mode wrote:
Hi Brian
May be Jhon can reverse the same ceramic filter and get a comparative response
with forward curve already succesfully published here , i expect to found
almost the same response, indeed the ceramic filter is loaded with same
virtual impedance of 430 Ohm on both sides.
Thanks John for sharing your superbe Z-renormalisation experiments.
73's Nizar .
|
Hi Brian,
I went ahead and tested the filter response using a 10 pF, 33 pF, and 68 pF load capacitor. Attached are 3 plots and the reference plot on each of the attached plots is when I'm not using a load capacitor.
It should be noted that the curves shown in the datasheets you attached have the left vertical axis identified as attenuation in dB. In reality it appears that axis has been normalized based on the peak of the response curve. In my attached plots I have not normalized the vertical axis so you can see what the true attenuation is through the filter.
Just FYI,
Don
|
Thanks, Don. Your filter shows much more response variation with a capacitive load than the 230 kHz filter in the Murata curves. I don't think anyone adds capacitance on purpose. In the past I have used pots instead of fixed 330 ohm resistors and adjusted them for minimum detected audio distortion. Since I now have quite a few spare filters, these days I just swap in another one, trying both orientations for minimum distortion.
Before I wrote this program, I used to worry a lot more about audio distortion:
Brian
|
Hi Brian,
I repeated the tests but this time using resistive matching pads to provide 330 ohm impedance to both the input and output of the ceramic filter versus using my very broadband transformers that provided a lower impedance (approximately 280 ohms). I also repeated using 270 ohm series resistors for matching. Since we now are only looking at the upper part of the response curve the loss in measurement dynamic range using the two different resistor methods should not be a problem.
Much different looking response using the resistive matching versus the transformers.
I again tested with a 10 pf, 33 pf, and 68 pf cap output loads and my reference curve used no capacitor on the load side.
Just FYI,
Don
|
Don, how about measuring the filter with a 10 pF load using no impedance matching (50 ohm source and load). Reverse the filter, measure again, and post the two .s2p files. I'm curious what my program will show dozens of dB down where the resistive pads limit the dynamic range.
Brian
|
Hi Brian,
Your wish is my command. See attached .s2p files for my ceramic filter with 10 pf load connected direct to NanoVNA-F ports tested forward and reverse connected.
Note: I set my NanoVNA Saver to 5 sweep segments to provide decent resolution and hope it's adequate for your needs.
Don
|
The plot compares S21 of the forward 50 ohm file with S21 after forward and reverse were renormalized to 330 ohms. Stopband details of the renormalized curve differ from your earlier curve using a transformer, but the passbands look similar. Thanks for all your help, Don. Also thanks to John Gord for explaining why renormalization works even with improper device loading, which I didn't expect.
Brian
|
