Introduction/Scope of Project
In light of a new 900 MHz hobby-within-a-hobby, I needed a sharp band-pass RF filter. The filter will be installed between an antenna on a radio tower and a small 900 MHz communications device. The receiver in the device (shown below) would normally not be capable of operating properly in a high RF environment without such a filter. The active device is a Heltec V3, a 150-milliwatt transceiver loaded with firmware such as Meshtastic or Meshcore. It is one of many types of off-grid messaging devices that are quickly gaining popularity. I will not elaborate further on the device or platform itself, as this is a hardware document concerning RF filtering. Pulling up a search engine and looking for Meshtastic or Meshcore will be the easiest way to learn about the platforms.
Shown here, a 150-milliwatt 910.525 MHz Heltec V3 transceiver
The Celwave (RFS) 800 MHz duplexer
About 10 years ago, these were extremely plentiful in the surplus world, so much that many of them were simply scrapped for mixed metal. They were primarily used for MDT (mobile data terminal) base stations, where the transmitter (usually a 100-watt Motorola Quantar) would continuously transmit data availability while the receiver listened to mobile field units. The duplexer allowed the use of one antenna and transmission line. Celwave (formerly Phelps Dodge), developed a genius one-box-fits-all solution without field techs needing to tune countless duplexers for every frequency imaginable. Using sharp gang filters, pass bands of 18-MHz wide for both transmit and receive were achieved while attenuation in the reject band (a 45-MHz split) was sufficient to allow proper duplexing. With an advertised loss of under 1 dB per side, many (and most) of the units sported much less loss than the specifications. Typically, each port only showed 0.5 dB of insertion loss.
The photo above shows the duplexer and the 19" rack panel it's attached to. This two-rack-unit panel allowed the Quantar and duplexer both to fit in one of Motorola's standard cabinets with no room to spare. With the disappearance of MDT's and most other site equipment sharing transmit combiners and receive multicouplers, these duplexers made their way out the door and unfortunately, the scrapyard.
Unit pre-check
If you are able to find one of these duplexers and you obviously have zero need for an 800 MHz public safety split and want to modify it for a 900-MHz filter (or filters--plural--maybe), then you will need to first do a pre-check to be certain the unit is operating properly and document pass responses BEFORE making any modifications. At minimum, you will need a two-port tracking generator that is accurate and will measure insertion loss. A one-port VNA will NOT work for this, although it is also nice to see the return loss or VSWR through each side of the duplexer to ensure that it matches the pass response with the two-port test.
In the screenshots above, the plots of the passband curve of the duplexer before any modifications were made.
Requirements - tools, experience and accessories
Good understanding of RF, filter responses and test equipment. As mentioned above, if you do not have a two-port pass response analyzer capable of going to 1 GHz, then stop here. A one-port VNA will NOT work for this conversion.
Good "shop etiquette" and experience in basic metalworking, cutting, filing, drilling, etc.
Clean workspace, patience, and several hours of time to do things right
Tools - drill, tubing cutter, pliers, rivet gun, screwdrivers, 7/64" Allen wrench
Tools - 5/16" wrench, 1/8" drill bit, file, deburring tool, etc.
Two gin-based cocktails (optional)
Initial setup with tracking generator. Be certain to terminate the unused port while sweeping with a 50-ohm dependable load.
Disconnect test equipment and set aside. Remove all plastic dust covers on the duplexer with a small screwdriver. These are delicate and easily broken, so pry gently.
Remove the 19" rack panel held on with Philips screws. Set screws aside. This will expose the tuning rods (front) of the duplexer.
Use a 7/64" Allen key to remove the screws in four locations. They aren't visible, so you will have to "feel" for the screws. Unscrew them completely and let the screws stay loose inside the duplexer for now. We will remove them later.
With a 1/8" bit, drill out ALL rivets. Be very graceful and careful not to use excessive force which may cause damage to the end plates.
Using a 3/32" nail set, carefully push rivet remains out of the way. They will fall into the duplexer cavity (no worries at this time)
Use the screws that once held the side brackets to the front panel to grab with pliers and gently remove the side panels from the duplexer case. This is the perfect time to dump rivet remains and any loose "stuff" out of the cavity.
The duplexer center conductor junction is soldered into place in the photo just where the cutter jaws are. It is only visible/accessible after the end plate has been removed. Cut this link, we will no longer need this junction.
The receive portion of the duplexer can now be separated from the transmit. The receive section is the smaller of the two and will have only ONE N-type connector. Set this aside, we will not be using this part of the duplexer in this document. It can and will be used as a second filter in another document (zero waste).
The main filter modification procedure can now begin.
Now that the two units have been separated, the last few rivets can be access and removed.
This is the other half of the center conductor that was clipped earlier. It is threaded into the coupling bar and can be unscrewed with pliers and discarded.
Coupling rods are secured with Phillips screws on the SAME side as the tuning slugs. Remove two screws, one on the left and one on the right.
Using a smaller (#1) Phillips screwdriver, remove the N-type flange connectors and pull them straight out.
Tuning rods are held in by Phillips screws on the same side as the N-flange connector ports. As you remove the screws, the rods will attempt to fall out. Use long nose pliers to reach in and hold the rods steady. Obviously, start from the ends and work your way to the middle.
Use long-nose pliers to reach in and hold/remove each tuning rod.
There are six tuning rods that determine the pass-band and ONE shorter tuning rod (outlined in yellow) that tunes a reject notch. Set aside the shorter rod, as it will NOT need to be modified. The pass-band rods measure 75 mm.
With a tubing cutter, trim each rod by approximately 3 mm, giving a total length of 72 mm (+/- 0.5 mm)
Use a deburring tool to "clean up" the rod ends. A step-bit works well for this.
Six of the rods should now be 72 mm in length. Note: two of the six rods have a small copper strip soldered to them. Leave the strip in-tact, as this determines coupling in and out of the filter.
Reassembly
Install the six modified tuning rods back into the cavity with long-nose pliers. With the filter in the position shown above, the SHORTER rod that was set aside in an earlier step should be installed in the position where the yellow box around the retaining screw is shown. This is beside the "TX" decal next to the N-type chassis connector. This is the NOTCH adjustment and will have very little effect on the pass band. We will get to this later on the service monitor screenshot.
After reassembly, before even touching the adjustment screws, check to see where the passband currently resides. Even without adjustment, we can visibly see the passband has been moved from the 855 MHz range to the 920 MHz range. This proves success.
SEE TEXT BELOW THIS PHOTO FIRST BEFORE ATTACHING END PLATES.
If you were careful with drilling the old rivets out, the new 1/8" rivets will attach perfectly. Use copper rivets if possible. If aluminum rivets must be used, apply a light coat of no-oxide grease to the end plates and the rivet heads.
Tuning Rod Orientation - READ FIRST BEFORE ATTACHING END PLATES
It would be an extremely good idea to attempt tuning the filter BEFORE attaching the end plates with rivets. The end plates should be loosely fashioned into the side of the filter as if you were preparing to rivet them in because capacitance to the end plates will affect tuning. Use tape to hold the end plates securely while tuning the filter. You may need to move/remove tuning rods (and therefore the coupling rod also) to make minor adjustments, such as a spacer/shim as I had to do on one of my coupling rods. If you rivet the end plates into place and run out of adjustment room on one of the tuning piers, you will ultimately have wasted rivets and time while drilling the rivets out (again) to access the inside of the filter.
I was also not able to get a good exclusive photo of the small copper strips soldered to two of the six tuning rods. You can see this on the rod in the photo below. These copper strips are capacitance adjustments and they MUST be mounted next to and facing the input/output coupling rods attached to the N-type chassis connectors. This gives maximum coupling and therefore less loss. More loss does not necessarily mean a sharper filter response in this case. Although this isn't critical, it does lower the insertion loss substantially if the copper strips on two of these tuning rods are located next to and rotated to face the N-flange connector coupling rods.
What happens if you cut a tuning rod too short?
If, while tuning the filter pass-band you find that you are running out of capacitance on a section (turning the screw clockwise until it bottoms out) then the rod is too short. A quick easy fix is to use a few small washers as a spacer where the rod is mounted into the cavity. This will effectively increase the rod length, therefore, increasing capacitance to the tuning slug and lowering the frequency for that section.
Tuning
With the tracking generator configured as before, attach the cables appropriately and begin tuning the filter. Adjust each screw alternatively. This will take time and patience, as each section of the filter affects another. Try to adjust the pass-band to appear like the photo below.
Try to adjust the filter to appear like the figure to the left. The pass-band should be less than 0.5 dB of loss. The notch adjustment (the adjustment next to the N-connector labeled "TX") can be adjusted to notch out troublesome frequencies. In my case, I adjusted the notch to 855 MHz.
If desired (and you should), check the VSWR through the filter into a known, dependable 50 ohm load. I get 1.3:1 throughout the entire passband which is excellent. I get extremely irritated at people who say "what difference does it make. It's just 150 milliwatts!" The effect that a 2:1 VSWR has on 150 milliwatts is the exact same ratio as a 2:1 VSWR has on a megawatt transmitter. Efficiency is best when impedances are matched, which is exactly what VSWR and return loss show. The amount of reflected power, in watts, will certainly be less with a one-watt transmitter versus a kilowatt transmitter--but the apparent and net effects in percentages of lost efficiency will be the same!
Update 3/6/2026 - How did it work?
I mounted my RF device in an aluminum enclosure with a 3D printed tray which was custom built by K4ROM. This mounted nicely to the filter itself, making a 19" rack-mountable all-in-one device.
The Meshcore repeater blends in very well with other commercial equipment at the site.
The antenna array on the left is a pair of DB408 UHF 450-470 MHz antennas. One antenna (the upper) is being used for a UHF repeater. The lower section is being used for Meshcore.
The antenna array
The antenna is mounted at the 500-foot level on a TV broadcast tower. The DB408 UHF antennas (and probably the DB404 and DB420 versions) seem to work fairly well on 910 MHz. It is nearly impossible to come up with a conclusion to why, but logically a half-wave on 450 MHz is a full-wave on 900 MHz. Does this give reason to why it works fairly well? There's no way to know unless range tests were done because the pattern is very unpredictable outside of the antenna's normal frequency range.
The filter in this article does an excellent job at keeping unwanted signals out of the Meshcore device. Right beside the antenna being used for Mesh is a powerful 10 dB gain 851 MHz antenna attached to a multi-channel trunked system. It has no effect on the Mesh device. Furthermore, the 20KW television station (561 MHz) along with a NOAA WX transmitter and two 100KW FM broadcast stations right across the street, the Mesh device performs flawlessly with noise floor of -118 dB according to the device itself. The device shows a -120 dB noise floor with no antenna attached. That is pretty outstanding.
Conclusion
I am very satisfied with the performance of the filter. I plan on converting the other half of the filter (the smaller portion) in the near future
Update 3/12/2026
The smaller of the two filters has been modified and tuned. The modification required cutting about 5 mm off each of the six tuning rods. However, this meant that the plungers (small brass threaded inserts) were too short to do any tuning. I picked up six #8-32 brass machine screws (stainless will also work) along with some #8 washers and nuts. The screws are 2" long, which is enough to tune each section effectively.
The 2nd filter is now complete. The end plates were riveted in like the previous section. There is plenty of tuning range left in the screws, and the center is again on 910.525 MHz. I duplicated an N-type chassis coupling rod, drilled and tapped holes, and mounted it on the far end.
The small filter actually surpsied me in overall performance. The rejection in the lower portion of the spectrum isn't quite as good as the larger filter, but the insertion loss is about the same. Perhaps the reason for the larger filter's size is due to power handling capability (it was formerly the TX part of the duplexer). I am extremely satisfied in the results and will be modifying more of these in the future as long as Meshcore/Meshtastic continues to be a useful thing
N4DBM