EMRFD Message Archive 13799

Message Date From Subject
13799 2017-04-13 18:44:42 lmeeny Affects of real component characteristics on BPF response.

I've been using the programs ttc.exe and gpla.exe to design several bandpass filters. After carefully measuring the values of inductors and capacitors I've compared the predicted to measured results taken by a vector network analyzer. In all cases the measured Q, center frequency divided by the 3dB bandwidth, is significantly higher than predicted results. What characteristic of real vs. ideal components could account for the difference?


13800 2017-04-13 18:53:44 Bill Carver Re: Affects of real component characteristics on BPF response.
What frequency are the filters for?

13801 2017-04-13 19:15:58 Dana Myers Re: Affects of real component characteristics on BPF response.
13802 2017-04-13 19:30:31 iam74@rocketmail.... Re: Affects of real component characteristics on BPF response.
When in doubt, first check your instruments of measurement and their true calibration.
Then, as noted, every result is frequency dependent.
If these are right, the most likely source of difference is unaccounted for phase or harmonic noise.

Lastly, who made sure the computer programs are correct and 100% accurate? How were
their instruments calibrated? (They had to check the results, No?)

Then, consider what are the real world characteristics of your components?

13803 2017-04-13 21:48:47 Bill Carver Re: Affects of real component characteristics on BPF response.
I babbled about bridges and instruments......and you have a VNA? Say a
nice HP box that goes to 1.3 GHz, hi?

Since you have access to a vector network analyzer do this: take one of
your components that you have measured on bridges or LC Meter or
whatever. Then put it on a connector and see what the vector network
analyzer tells you it is at the bridge frequency, then at the frequency
of the filter you want to build. Are they the same? If not you can see
the effect of "strays" and evaluate, for yourself and your situation,
what you have to do to make that component better, or modify it to at
least "look like" the value your filter demands at the filter frequency.

Bill W7AAZ
13804 2017-04-13 22:05:27 Bill Carver Re: Affects of real component characteristics on BPF response.
Email is a difficult medium to zero in on solution to something because
it takes "forever" to go back and forth. And there's not enough
information in the original post of "filter not matching the design",
although Dana offered a response that points out the "strays" in
components can be the issue.

I think it is extremely unlikely a reputable computer design program
(Elsie, BAND, etc) is faulty. But any program can be asked to crank out
a filter that is physically difficult to build with real-world
components. For example in a 50 ohm system a program might calculate 80m
inductors that are 0.1 uH, resonated with .02 uF capacitors. In such a
case winding turns of magnet wire around a pencil will make an 0.1uH
inductor, and a .02 uF 400V paper capacitor might be in the junkbox, but
it still won't work. Programs like Elsie gives you the range of
inductances in a particular filter....getting above 10:1 in a single
filter is an indication you might have problems.

When that happens you might change the impedance level of the filter,
transforming it back to 50 ohms with wideband transformers at the end.
Or put taps on shunt inductors, or change the filter topology to one
that gives nicer component values. That is, if you get extreme component
values with a "pi" type filter, telling the program to compute the same
thing with a "T" toplogy might yield L and C values that aren't so extreme.

I have "oldie but goodie" Boonton vacuum tube L and C bridges and an
HP4271B 1 MHz digital bridge, as well as a homebrew LC meter and an AADE
LC meter. They are in very close agreement......at 1 MHz. And if you're
making an 80m filter and the component values aren't "too extreme", the
1 MHz measurements work fine. But as Dana says, EVERY component has
strays and as the frequency rises their effective value changes,
particularly inductors. Another "oldie but goodie" instrument is a
Boonton 260A Q meter: it isn't anywhere near as accurate as the bridges,
but you can put on your old-time hat and spend an afternoon making
measurements at several frequencies to compute the stray capacitance of
an inductor, the series inductance and resistance of a capacitor, and
maybe compensate to make them act like what's called for at the filter

Another technique is to use the "Dishal" method to tune the filter to
the desired center frequency. You'll need to look that up on internet,
but briefly: you open circuit and short circuit places in the filter and
tune that place for highest voltage or a null at the input/output ends
of the filter. Tedious, I don't even know how to apply it to some filter
types, but I have used it a few times. It does have the advantage that
you just need an accurate signal generator frequency rather than a bunch
of fancy test equipment.

To take advantage of the experience of others, we'd ideally like to see
the filter schematic with component values on it and the impedance level
defined, and perhaps a brief description of how you measured it and how
far off the results were from what was predicted. I've built an awful
lot of filters over the years, avoiding designs with extreme component
values, and essentially I've never had a filter that wasn't very close
to the predicted response.

Bill W7AAZ
13805 2017-04-14 11:07:01 Dana Myers Re: Affects of real component characteristics on BPF response.
13806 2017-04-14 11:17:21 Dana Myers Re: Affects of real component characteristics on BPF response.
13807 2017-04-14 11:25:51 Graham / KE9H Re: Affects of real component characteristics on BPF response.
Ferrite core inductors walk all over the place with frequency.
You need to measure them at the frequency you plan to use them.

The little ADE type meters measure down below 100 kHz, which make them useless for HF applications

Iron core inductors are much better behaved, but still recommend measuring on the frequency of use.

Your VNA should work fine for that.

I use Elsie for modeling, and when actual on-frequency inductance measurements and actual "Q" values are used, agreement with filter as built is excellent.

--- Graham / KE9H

13808 2017-04-15 19:13:28 kb1gmx Re: Affects of real component characteristics on BPF response.
I use an AADE LC-II into the UHF region with success.  True the error does increase with frequency but not so severely that a 3 to 4 digit answer is not useful.  That assumes calibration is also valid.  

At HF its not as much an issue as getting close to start with.  MOSt circuits are not that narrowly tuned that its not workable.

Ferrite at RF is a bit of a beast and not usually well suited for filters or other tuned circuits though the FTxx-61 is not too bad at low HF (under 10mhz).  THe -43 material is fine for 
transformers and chokes but far from optimum for tuned circuits, too lossy.

Cores in the -2 and -6 are iron powder and are more stable with temperature and frequency.

FYI: if you hand the same core to three different people the resulting 10 turns will
have three different inductances due to tension and spacing.  The fix if tuning is 
needed is to move the wires around.  Or provide a tuning trimmer capacitor.

Or if the filter is of the low pass or high pass form not worry so much as the 
corners are usually far enough away.


13810 2017-04-17 12:56:34 lmeeny Re: Affects of real component characteristics on BPF response.

I took again all pertinent measurements on the as-built filters and recorded the predicted results.

Here are the results of the as-built filters.

                                                20M      30M     40M    80M
f (-3dB) MEASURED            13.58    9.76    6.81    3.45
f (center) MEASURED          14.10    9.89    7.06    3.58
f (+3dB) MEASURED           14.77    10.19  7.32    3.87
Q (MEASURED)                   11.86    22.58    13.81    8.58

Here are the results of the gpla.exe simulations

f (-3dB) PREDICTED           13.69     9.89     6.99     3.40
f (center) PREDICTED         14.19    10.09    7.15     3.63
f (+3dB) PREDICTED          14.77    10.32    7.41     3.88
Q PREDICTED                      13.15    23.25   17.07   7.57

There's quite a good match between the two results, well within the measurement accuracy.

Comparison of predicted vs. measure insertion loss was problematic. Here's the data.

                                            20M    30M       40M    80M
IL MEASURED                 1.57     2.62       1.85     2.57
IL PREDICTED                 1.20     1.93       0.17     0.09

I'll chase this one myself before bugging the group.               

I apologize if anyone feels their time was wasted. This has been a great learning experience for me.

The filters were tuned by squeezing or expanding the toroid inductors as required.

Plots of the measured filter performance are available should anyone be interested.

Thank you all very much.

13811 2017-04-17 20:11:25 Ken Chase Re: Affects of real component characteristics on BPF response.
Hi Ed

Nice work on the filter data and it's not a waste of others time. I do appreciate it.

What filters are you using?



13812 2017-04-17 21:45:35 Bill Carver Re: Affects of real component characteristics on BPF response.
I still don't know what kind of filter this was: how many poles, etc, but for hip shooting I'll assume it was a 2 pole Butterworth filter.

I used Elsie to design a 20m filter assuming a reasonable value of 2000 for Q of silvered mica capacitors. The Q of the inductor would be specified as 145 to get that 1.2 dB of insertion loss. That Q is easily obtained at 14 MHz. But the actual bandwidth was wider that you expected and the insertion loss was a bit higher. So I again designed a 2 pole Chebychev for that width and insertion loss, and came up with a coil Q of 93. So I can hipshoot and surmise that actual 20m inductor Q was a bit lower than what you used for the design.

I did the same for the 40m filter. For the desired -3 dB frequencies I used capacitor Q of 20,000 (essentially infinite) and adjusted inductor Q to get 0.17 dB insertion loss. Getting that insertion loss requires an inductor Q of 1300: if not impossible, damned close to it. A real inductor will have a Q of perhaps 300 at the best. So my hipshot is you typed in one too many zeros for the inductor Q    ??

For the actual 40m 3 dB frequencies, still assuming lossless capacitors, I adjusted inductor design Q to get the measured 1.85 dB of insertion loss and came up with a Q of 94. Don't know anything about the construction of the inductor, but that's within the realm of possibility.

Going back to the first sentence: I don't know what kind of filter was intended: how many poles, etc. But hip shooting I'd say you put in the wrong value for inductor Q when you did the design for the 40m and 80m filters. And the 20m filter needs just a bit more Q in the coils. If you don't have a Q meter or are able to get a Q measurement from the network analyzer, you could get on the Amidon Website and look at the Q v.s. frequency for different toroidal coil designs that Bill Amidon compiled many decades ago.

Putting my gun back in the holster, this was my attempt to understand what might be going on. If you say "Nah, these are 3 pole filters and they're Chebychev", OK. The design simulations will come out with different numbers. But I can guarantee they will still come up with impossibly large numbers for 40m and 80m coil Q. So you might want to go back and look at the designs again and see if any of these comments make sense.

Using reasonably accurate coil Q, or guesses from the Bill Amidon's examples, and measuring L and C with something like the AADE meter, you should be able to duplicate 80m and 40m designs within a few percent on -3 dB frequencies and pretty close on insertion loss. 20m is slightly harder because stray capacitance becomes a larger percentage of the total capacitance, so you might subtract a few pF from the calculated values when you build the filters for 20m.


13813 2017-04-17 21:56:42 lmeeny Re: Affects of real component characteristics on BPF response.

I decided to use the classic triple tuned capacitor input bandpass filters given the support of the design by the program TTC08.exe supplied with the EMFRD included software.

An early lesson was the relationship between filter Q and hi-side rejection. On 80M particularly I designed for a wide filter resulting in very low Q and poor hi-side rejection. If I had it to do over again I'd select an inductor input filter. I'm thinking though that AM broadcast stations are the worst offenders on 80M where the rejection of the classic filter is good on the low side.

Another lesson was that as the Q of the filter increases so does the insertion loss. I'd not made that connection before.

In troubleshooting I learned that if the center frequency of the filter is correct but insertion loss is horrendous you haven't soldered both ends of a parallel inductor into the circuit. HiHi

All in all an occasionally frustrating  but educational valuable experience.