Crystal Filters ~ The Latest Craze, DIY.
Updated 1/8/2022
This post is prompted by my receiving the latest SPRAT #189. The 1st article is by G3OOU who has an excellent piece on the design of Crystal Filters. A really great read. So, if you are not a G-QRP Club member --you won't see it.
A view of crystal filters from 10,000 feet reflects that a crystal filter has several basic foundational elements inclusive of the crystal themselves, the filter capacitors and the matching transformers. How these are arranged, and their specific values impact the filter's response curve. The number of crystals used in the filter directly impacts the filter skirts. This means how close to vertical the slope of the crystal curve (the brick wall filter look). Thus, more crystals the tighter the skirts! Graphically the bandwidth at -60 dB down is narrower with a tighter filter. You ask why is that desirable -- the answer is rejection of close in signals to the sides of the one being copied (brick wall).
Many homebrewers for SSB applications start with a 4-pole filter. But soon the lust for signal rejection from the side (those wide skirts) will drive the journey to a 6, 8 or even 10 pole filters. The journey to more crystals in the filter just takes time and a lot of fiddling to have everything "Tickety Pooh".
The capacitors on the other hand determine the filter bandwidth with smaller values resulting in WIDER filter bandwidths and larger capacitors having the effect of narrowing the bandwidth.
Meanwhile the matching transformers directly impact the pass band ripple. The flatter the top of the response curve the more desirable the filter insofar as signals passing through the filter with a constant amplitude.
Yes, it is a highwire balancing act as ahead of any parts soldering is the Crystal Characterization where the homebrewer must know firstly the exact frequency of the crystals being used so that across all the crystals that the frequency spread may be as low as 50 Hz total.
TKT (Tribal Knowledge Tip) Most successful filter builders purchase a large number of crystals as a starting place (like 100 at a time) so that to find 4, 6,8 or even 10 crystals closely matched takes a large starting pot. Likely you will end up with several piles of candidates that meet the 50 Hz specification; but you had to start with a large batch to find the matching crystals.
N6NWP cautions about buying a small batch first as a starting place and to derive the Q factor for that particular crystal BEFORE Jumping in the Pool with the 100 Piece buy that might cost you $50.
The homebrewer will also need to know the motional capacitance, effective series resistance and the "Q" of the crystals. Some of these values are easily derived, some come from the manufacturers published specifications and some require divine intervention to find.
The reason for the data gathering is that there are canned computer programs that will generate the values of the capacitors and impedance matching and do so depending how many crystals are in the filter. You have the AADE Program, one from W7ZOI and of course Dishal (Which I call Dishal Dystopia). Let us not forget the Program specified in the G3OOU article.
For the "noob" you will soon learn about the G3UUR oscillator circuit that tests crystals and gives two of the values you need. There are a cast of players that offer resources on how to successfully build Crystal Filters such as G3UUR, W7ZOI, ZL2CTM, WA5BDU and N6NWP.
I took a stab at brewing up a 4-pole Crystal Filter using some crystals I had long ago stashed in a box marked "Win the Lottery." I often find in building a filter my personal chance for success the first time is actually less than winning the $500M Power Ball Lottery. Note I mark the crystals with a number and enter that on an excel spreadsheet, so I know which is which.
After some "Diddling" with the matching transformers, here is the response curve of the filter above.
Perhaps not the final form but a place to start and that I did. I called up the Nano Saver app and found that this 1st cut was rather narrow, at only about 1.6 kHz at the 3dB points which meant I needed to reduce some of the capacitance values. The BW is now 2.1 kHz and so that will work.
But as you change things not just one thing changes -- now the ripple is not as good as it was at 1.6kHz. So, more tweaking of the turns-ratios. to match impedances
I suspect the heightened interest in Crystal Filters is the result of the Nano VNA and the ability to actually look at how the filter performs. That said if you are contemplating a DIY Crystal Filter, cruise by this you tube video from ZL2CTM, Charlie Morris. He demonstrates how to build a superb filter without using a Nano VNA.
Jacob, N6NWP has authored several excellent articles on "brewing up" a DYI Crystal Filter. See the link below.
One very important point in the N6NWP link is the section dealing with Frequency Selection. N6QW's axiom about "where you stick it really matters" shines through like a beacon!
The sticking it issue has at least three concerns:
1. The possibility of unwanted mixing products.
2. Using cheap crystals and their frequency stability
3. The LO frequency ~ Above or Below
I have exceptionally good experience using 4.9152 MHz crystals in a homebrew filter. I tripped over this frequency when I read the Elecraft K2 product description in 2006. Good enough for Eric, good enough for me. Jacob suggests that any filter selection frequency below 6 MHz is not a good idea --so a bit of a conflict in the K2 and N6NWP on the filter frequency.
Now today with the Arduino and Si5351 literally any Filter Center Frequency is not a problem with the BFO frequencies or the LO. It is just a matter of what numbers you stick in the sketch.
But suppose you didn't use the Arduino and the Si5351 but instead used a VXO then again "where you stick it matters". Going back to my 4.9152 filter this again was a good choice for me and here is why.
I like super VXO's as you get more than just a few kHz of frequency swing AND there are cheap computer crystals that work well with the 4.9152 MHz Filter.
The first is 20 Meters and there are computer crystals available at 19.2 Megahertz. Put two of those in a super VXO and you get about 30 kHz of stable frequency spread. I have built such a VXO so that 19.2 - 4.9152 = 14.284. In practice you find the resultant frequency is a bit less than that and typically tunes downward. [Lew McCoy wrote a VXO article in QST using a Dual Gate MOSFET where he was able to tune upward]. So, my 1st shot at this had a VXO that let me tune 14.276 to 14.244 MHz which puts a rig in a nice part of the 20M Band.
Using a higher VXO Frequency Crystal enables a greater frequency swing. Now a subset to this approach is to use a crystal oscillator at 6 Megahertz mated with a NE602 as a 12.96 MHz VXO + Mixer where you can generate 19 MHz signals.
But the twist is that I have several 6 MHz crystals of different frequencies that are switch selected from the front panel so in essence there are several VXO frequency ranges that can be called up all within 20 Meters. The smaller of my Shirt Pocket Transceivers used this approach to cover about 60 kHz of 20 Meters.
Now the next band of choice, 17M is where the 4.9152 Filter earns its stripes as you avoid some of the problems with using a 9 MHz filter on 17M.
This idea was passed on to me by Jim Kortge K8IQY. One of the available computer crystals is 11.52 Megahertz which of course by itself does nothing for you. However, if you run that through a diode doubler circuit the output is 23.04 MHz where 23.04 - 4.9152 = 18.1248 MHz which of course is in the nice part of 17M. Here is the bonus Jim shared with me. Whatever you are able to swing at 11.52 MHz is doubled at 23 MHz and the doubler requires no power!
This filter frequency coupled with a non-Arduino approach opens the door to many homebrewer enthusiasts who fear the Arduino but want to brew something. The smaller size of the VXO also suggests opportunities for more compact gear such as you might want on a SOTA or POTA foray. If you are able to swing down to 18.1 MHz, then you are set up for FT-8.
Yes, placing the LO above the working frequency causes sideband inversion. The reference to The Gout -- you ate the wrong things during the Christmas Holiday never ending party!
This is fun!
73's
Pete N6QW