simple 15mhz crystal filter how to?
I will use it as a narrow BPF on my spectrum analyzer project, after a Gilbert cell mixer.
I thought a single series crystal would do the job, but it seems id does not, it just attenuates a bit frequencies out of resonance, but the performance is bad.
Will such a filter work ok http://www.qsl.net/m0ayf/QRSS-Band-Pass-Filters.html (the second picture)
I think it should works. Could you, please, post the schematic ?
The schematic of my spectrum analyzer is at
www.qrp.gr
(find in this page the link "Poor man's HF spectrum analyzer" near the end of the page).
The schematic of the crystal filter I have found is this one.
[found at: www . qsl . net /m0ayf/QRSS-BPF/Xtal-Filter2.gif]
Now I use just a series crystal and it does not work too good.
That should certainly work, its the well known standard half lattice filter.
The problem you might find with this is filter ringing, the sweep speed of your spectrum analyser must be made suitably slow. The usual application of these is for receiving CW signals which have a relatively slow on/off rate, only a very few Hz at maximum. Listening to audio through such a filter sounds quite strange.
Its not usually possible to get a very high out of band attenuation with just a simple single stage filter, especially at a high operating frequency. Layout and screening is absolutely critical, but its going to be so much more difficult at 15 Mhz than at 500 Khz or even lower.
The usual approach is to cascade several filters with gain stages in between having progressively smaller bandwidth. The last and narrowest filter sets the overall bandwidth, but the earlier broader filters greatly attenuate the out of band signals.
The half lattice filter in a serious CW receiver is usually placed at the last IF stage, where earlier IF amplifier stages have already greatly attenuated the wider out of band signals. It just adds a very narrow peak to finish the job.
All right, I suspected it.
I thought to include an LC BPF after the mixer and before the crystal filter. This should reduce quite much the "out of band" signals. Then the crystal filter will pick up a narrow portion of this signal and have to do "much less work".
I insist in using a single crystal because it needs no crystal matching (difficult for most), so it does the job easier for the homebrewer. The sweep rate, if one wants narrow bandwidth, must be low anyway My Tek 491 has a sweep rate of 1second to give about 1-3KHz resolution.
A problem I have is harmonics of the VCO, with the approach I used. I think I should better use a higher frequency vco, above the frequency of coverage, so all the harmonics can be filtered by a LPF. But then this would change the whole analyzer block diagram and also, I should try to search for an IF filter higher than 30MHz which is difficult. Even the TV IF ceramic resonators have a few KHz bandwidth. Nothing beats the single quartz crystal filter! So for great resolution the only solution in a single conversion analyzer, is a crystal filter.
That would probably be your best chance with a single conversion analyser.
Leakage around the filters will be your worst enemy.
LC circuits are not that selective, a Q of a hundred might be about the best you can hope for with a -6db bandwidth of 150 Khz at 15 Mhz. So its going to need several tuned circuits well shielded from each other, but its possible.
Commercial multi stage crystal filters are readily available for 45 Mhz and 70 Mhz with typically 15 Khz bandwidth and pretty steep skirts. One of those might be worth a look.
I am chasing VCO harmonics myself right now. A pretty brutal low pass filter on the output is the usual answer.
But another is to build a reasonably good sine wave oscillator with low harmonic distortion, so there are minimum harmonics generated at the source. It takes far less cleaning up if the third harmonic is fairly minimal straight out of the oscillator.
A LPF is a solution for a narrow range vco. For a wide range, such as the one used in this analyzer, this is not a good choice.
I agree about the low harmonics content oscillator, I have designed a vfo that has harmonics -35 to -40dbc, but It cannot cover the frequencies below 1MHz or so and I do not know how well it would behave by the addition of a varicap. Experimenting more is needed, but for the time being I am trying to make a usable analyzer out of discrete components, so I will stick with this wideband vco.
But I am thinking, would the harmonics of the LOSC be so much of a problem? I mean could I identify them in the display so that they are not confused with the usable signals? Maybe they are moved in more speed onto the display than the signals, and in different directions.
I will try the filtering you propose. Do you think an amplifier is needed between them? I thought not to include one.
If you are putting lots of signal in the crystal input, it has nowhere to dump the energy of the out-of-band frequencies. You must ensure that the signal input energy is low and the output is also low. In other words, you must maintain a high Q for the whole system. That mean that there must be other RC filters on both sides of the crystal.
It depends on the actual frequencies and filtering involved.
If there are narrow filters both ahead and after a mixer, spurious mixer problems and harmonics are not usually a problem.
Only the good stuff makes it through both filters, everything else just gets lost.
Its where you have truly broad band input straight into a mixer where you can get into real trouble with spurious mixer products falling within the IF filter bandwidth.
Also a lot depends if you are up converting to the IF, or down converting to the IF.
The cleanest way to go about all this is to up convert to a ridiculously high first IF, then down convert in several stages to some ridiculously low final IF.
All very complicated and expensive, but its a well proven path to success.
One thing is for sure.
After attempting to home brew this type of thing, it leaves me in absolute awe of the performance of well built professional grade test equipment
I would also like to mention the contribution of mathematicians in electronic engineering. I do not understand the details and I do not know how it is done. The wonder is that someone finally translates the strange equations (and inequalities) into capacitors and resistors.
why in both sides? Why an LC filter is needed at the output?
Ok here is the filter I have designed.
The 47pf must be within 1% or else the frequency top will vary. The 2.2uH must be quite accurate too, but if using T30-12 I think it will be.
The other two shunt inductors set the steepness of the filter, but I set them quite wide. After all this is a pre-crystal filter.
That must act as the sink; just to prevent reflection back into the other side...
The maths is certainly interesting, but far more important is a basic understanding of what the whole purpose of the thing is, and the causes and solutions to unexpected problems that arise.
A mathematician can probably design some super mathematically pure filter that has 200 db of rejection, but cannot be made to work in practice, because even if the magic filter is totally removed, there is less than 200 db of attenuation between input and output due to stray coupling.
Theory is fine, but it needs to be tempered with some practical experience.
It is legendary that some circuit designed on a simulator does not work when built.
Grounding, bypassing, layout choice of suitable capacitor types, can be just as important as the actual schematic and component values.
The higher the operating frequency, the more important the practical "hands on" side of circuit design becomes.
I agree with you.
So in the practical part, I have made the filter posted. I have used 37T on T30-12 fir 2.2uH. Because of the low permeability it needs to be 37T, which of course increases the precision headroom, because the more turns, the more precise you can tune the inductor (steps).
T30-12 was used for the other two inductors of 100nH. This is not of great precision, but the value seems to more vary the width, rather than the frequency top too much.
The capacitor is a silver mica +/-1%. It needs to be accurate, else a variable capacitor must be used. However, then tuning would be needed and also the temp. coeff of the variable capacitors is usually bad. +-1pF really changes the top quite a lot.
Now the tests results were satisfactory!
The filter has been placed after the mixer and the products were attenuated. I cannot say how much, I did not do any measurements. However, when I connected a series 15MHz crystal after the filter, almost all products were eliminated.
The only products I could see on the FFT, were the signal source (mixer RF) and some 15MHz signals that were coming through instantly and then disappeared. This was due to the sweeper that sweeped the local oscillator (and it's harmonics). When any of these signals were passing from the 15MHz crystal filter and the previously connected BPF, it appeared at the FFT at 15MHz instantly, then disappeared, because of the sweeping action.
How far of the top of the BPF from the optimum designed point is (due to filter components variations), only adds loss to the system. It seems that your proposition and the filter I made, work ok. However I have not measured the loss.
That all sounds really good, and its very satisfying when things work out well.
I bought three 70 Mhz surface acoustic wave filters (1 Mhz bandwidth) from e-bay and hooked one of them up to my spectrum analyser and tracking generator, and could hardly see any effect at all of the filter.
It was supposed to have something like 70 db rejection and 1db ripple.
My immediate thought was that I had been shafted, sold three faulty reject filters....
These were specified as being 50 ohms in and out impedance, (same as the spectrum analyser), and I used 50 ohm coax with about 25mm stripped at the end and soldered onto the filter pins at each end.
What could possibly go wrong ?
After trying all three filters, turning them around and trying all sorts of things I eventually discovered that 25mm of exposed wire at each end created enough leakage around the filter to completely destroy any filtering effect.
By using ultra short connections of only a couple of exposed mm between coax inner, and filter pin it started to get better. A copper screen soldered across the filter can between input side and output side fixed it.
The effect of the copper screen really surprised me, because the filter in and out connections are 20mm apart on a steel can.
The response then looked exactly like that on the data sheet.
It was all a lot more critical than I ever expected it could be.
But it was a lesson worth learning.
Not to underestimate the effects of stray coupling. And if really high out of band rejection is required, its much better to spread it between more than one cascaded filter stage. Its a lot easier to get 35db rejection twice with two well separated stages, than try for 70db all in in one go.
Thanks, all this information was very helpful and confirms my tests. At 15MHz stray coupling is not that critical as in 70MHz though, but because +/-1pf at the capacitor does a difference in this "critical" designed filter, I tried to minimize the coupling between the components in the small proto-board, in which the filter was soldered. I left some in-between pins unconnected in the I/O pin row header, so that the input and the output are physically more far apart.
It seems to work fine, I will do more tests though.
Its probably a lot easier to get good results with a proper PCB with short tracks, ground plane and some surface mount ultra small capacitors than with a "birds nest" three dimensional prototype.
I have only had this spectrum analyser for a few weeks and it has opened up a whole new world of discovery.
Yes these are very usable instruments. You don't even need a scope most of the time for light RF work, if you have a spectrum analyzer. You do not need to "see" the sinewave to judge the quality of the signal, you just measure it's harmonics.
However these are way too expensive for the experimenter, so a few try to make their own, definitely not the best choice, but better than nothing.
My method is not the best, as usually high frequency local oscillators are used to cover the whole band of interest, but it saves a lot of money and pain from difficult to find filters and at the same time it should achieve high resolution, enough to see the sidebands of an AM signal I hope.
A spectrum analyzer is useless if you do not have a way to make some accurate as possible measurements.
For the measurements, I am thinking of using a second calibrated oscillator, fed at the input of the analyzer. The oscillator will be tuned at the frequency of the Rf signal to be mesured and it will be used as a marker, so that the frequency of the signal can be measured, but also the amplitude.
Any losses in the spectrum analyzer chain, would apply both to the RF input signal to be measured, but equally to the callibrated oscillator, so we should normally do not care about these losses, as we just compare the two signals.
A similar technique was used in some Tektronix analyzers at the past, but they used a negative pulse in the display, instead of another signal. However this could only measure frequency, not amplitude (https://www.youtube.com/watch?v=D3O2xIL-91g)
I am not so sure about that these days.
I just bought an e-bay Advantest 3361 spectrum analyser 9Khz to 2.6 Ghz with tracking generator for S650 US. These cost S75,000 brand new twenty years ago.
Government departments around the world are writing them off in large numbers, and second-hand dealers are selling some of the less lovely looking ones cheap on e-bay.
A pristine one might go for about S1,600 US. That's well out of my price range.
The reason it was cheap was the CRT had a scan burn, and the CRT emission is down to the point where its just usable at flat out full brightness.
I have ordered a new CRT for less than S200 US which should make the whole thing work as good as new.
I have lusted after something like this for a very long time.
Only now have they fallen to within my price range.
Thanks Tony,
My best shot to date is with a HP 54520A bought for about 250S. Not an SA but the FFT is much better than no SA at all.
The other thing I have is a Tek 491 which I bought at almost S500, but with all it's accessories to 40GHz. Not very "serious" these days but it is usable if you have a means of doing relative measurements. For a 40GHz SA one would pay SSSS today even used.
However, the circuit I am making is a hobbyist version with a 30MHz max frequency. I hope it will be usable with the addition of the marker oscillator I mentioned earlier...
BTW, shall I use a transformer after the LC BPF and before the crystal (single crystal half latice) or just hook up a series crystal (single crystal ladder)?
Now that I am thinking of it a bit more, I think There is a better way to achieve narrow BPF. See at the middle of this page http://lucafusari.altervista.org/pag...wabenland.html the crystal filter. Two simple resonant circuits and a crystal within. This satisfies the filter before and after the crystal, that was mentioned in this thread. Tuning shouldn't be that difficult.