5-pole Comb filter (cavity filter)

[Janos Konya]

This will not be about Comb filtering like it is most commonly used in audio technology - here the term refers only to the location of the poles. In this filter, the poles are on the same side of the box , not on the alternating side (Interdigital)

Learning from the mechanical problems of my recent experimental Interdigital filter, I discarded the idea of using thin materials. I chose aluminum as the material for the next project. It is cheaper and easier to obtain than copper, but its radio frequency properties are slightly inferior to copper. On the other hand, copper oxide has a worse effect than alu oxid on the skin effect of the radio waves ... So the decision was made.

While I waited for the materials and the missing tools to arrive, I played a bit with the online calculator of the Interdigital Filter - mentioned in my previous post.
I noticed that for the previously used dimensions, the impedance difference between the outer and inner poles disappears if the end plates are at least 35-40 mm from the outer poles. I definitely wanted to use the discovered phenomenon...

The filter body is made of 60x40 mm, 2 mm wall thick alu hollow section. For the filter, this is a little smaller than ideal, but the next standard size would be too large. The poles were made of 10 mm thick aluminum rods.

The length of the poles is 51.7 mm, centerpoints are 50 mm apart. The use of rods shorter than desired is limited by the internal dimensions of the "box", so a diameter of 10-11 mm is justified so that I can achieve a larger capacity in proportion to the surface so that I do not have to insert too much "iron"- (sic!) tuning screws into the box.
​

I could not find smaller than M5x16 (mm) screws at home, so I used them...

I got this far when I found an example on the internet that someone else had been on my way before. I took the distance of the probes from him with a minor adjustment.

link: Practical cavity filters (my browser gives safety warnings here, but ignoring the warning had not caused damages in my comp. Be careful anyway.)

My probes are 1 mm longer (41 mm) and 1.5 mm thick. I could not solder a thick wire directly into the connector, so I soldered a short piece 1 mm thick into the internal contact of the connector and cut it very short. I soldered the probe overlapping a small part, close to the base of the connector. Avoiding making a lot of tiny holes for the small screws used to secure the connector, - I chose a crimp connector that I cut off the crimp part of and sanded the hex to thin. (presence of excess metal inside the box is really unwanted) I secured the SMA female connector from the outside with the nut and washer originally obtained for the connector. Be sure to use instant glue - on the soldered side - to secure the center contact in place, otherwise the connecting pin may push the center tubular contact into the box. ... Without a crimped cable, nothing else would stop it.
Since the probe is located somewhat eccentrically, I drilled the connector hole 1 mm below the centerline of the box.
Extra feature: - By turning the connector, the coupling of the output and input can also be tuned without moving significantly from the center line ..

The length of the box was given due to the 50 mm between the center line of the 5 poles, the probes placed (2x)20 mm after the outer poles, and the previously decided 40 mm end plate distance twice. The size is 320 mm.
​

The finished product:

I tested the ends covered by alu and copper plates and also with open end - without experiencing real differences in measurement, so I used the non-metal end-plugs I bought with the hollow section.

By making a compromise between bandwidth and insertion loss, I chose the bandwidth between -6 dB points to 17 MHz. The filter's insertion loss does not reach 1 dB.

I can only estimate the attenuation from the filter characteristics because the dynamic range of the nanoVNA-H is worse than 60 dB on this frequency - and the lower part of the drawing is noisy. The parameters of the 5-pole filter exceed the available measuring range. I estimate that the near GSM spectrum is attenuated by around 80-85 dB, perhaps more.

After a short test, compared to the factory made 40-50 dB ceramic filter (SF9179):
- The number of messages per second increased by an additional 15+ percent and I got rid of the original 4 dB insertion loss of the mentioned filter..

Good news: - You can create this filter using your mechanical experiences only.
The "bad" thing: - VNA or nanoVNA is a minimal requrement to tune it properly.
By the way, a nanoVNA is not so expensive. My next one will be the V2 version. It has a 70 dB dynamic range...​

That's all for now...
Janos

 

[jms]

That is a big filter was difficult to tune it? My only experience was tuning a 3 pole filter using a noise source and the RTL-SDR dongle, easy task.

 

[Janos Konya]

Unfortunately, when using a noise source and receiver, there is not much to find for tuning - just the frequency. Better than nothing, but both devices are characterized by non-linear signal output and sensitivity. Basic calibration is not possible either. Of course, if there were no other means, one would be happy for them as well. Anyway, good job!

I did the tuning twice. At first, I was in too much of a hurry to tune it, like a kid unpacking his Christmas present. For the second time, the result was much better. In addition to setting the attenuation parameters of the filter, I also made sure that the desired values were achieved meanwhile the output and input were 50 ohms at the same time. I played with it for almost 2 hours. I had to take care of insertion loss, group delay, bandwidth, return loss (from both side) and Q factor for getting an optimized result. For example, dealing with fine tuning sometimes easier if I watch more than one drawings simultaneously. S21 gain diagram alone helps less than return loss or SWR and Smith chart together to minimize the ripple - at the finish line.
If I had made a 7-pole filter, I would definitely have used a U-shaped resonator chamber to save space.