The Final Filter Shootout

@abcd567 asked this week for a more complete one to one test of different filters. so - here we go:
a picture of what will be tested - the results follow as soon as finished.

HERE WE GO - MY RESULTS:

the site is located on the country side about 25 miles south of munich/germany. all tests on the same jetvision a3 antenna, rtl-sdr dongle and raspberry pi2. while test ran always about 200 aircrafts were in receiver range. pi + dongle used ran for 3 hours before test started to have same temperature from start. gain was fix set to 50. if you watch the scans not only look for noise but also for 1090mhz signal - because it’s all about signal-noise-ratio (snr).

  • for most sites the uputronics saw-filter/lna is the best bang for the buck
  • if noise at your site is somewhat heavier over the full range the uputronics ceramic-filter/lna is the best chioce
  • for those with really heavy noise over the whole frequencies a cavity filter in front assisted with the uputronic lna with ceramic filter behind is the way to go. the jetvision cavity filter has extreme low passband attenuation (0.5db) - that’s why this is the only filter to use in front of an lna
  • the flightaware filter has excellent attenuation far below and above 1090mhz - but unfortunately totally fails at all frequencies very close to 1090mhz and those are most often the biggest challenge for the receiver

edit: one thing i forgot to mention is that using the cavity (3-pole) filter some aircrafts behind the alps i get longer distance messages from. maybe this is because the passband attenuation is extremely low while mostly all noise is filtered out - and because of this the receiver (dongle) has a bigger chance to find and decode those faint signals …

you can plot your own scans following these steps:

install needed software:

sudo apt-get update
sudo apt-get install git build-essential cmake libusb-1.0-0-dev python-imaging
git clone https://github.com/keenerd/rtl-sdr.git
mkdir rtl-sdr/build
cd rtl-sdr/build
cmake ..
make rtl_power
wget https://raw.githubusercontent.com/keenerd/rtl-sdr-misc/master/heatmap/heatmap.py
chmod +x heatmap.py

now anything you need is installed and you can e.g. do those two scans:

sudo killall dump1090-mutability
./src/rtl_power -f 800M:1400M:300K -i 10 -c 50% -g 50 -e 20m -F 9 >scan_800_1400.csv
./src/rtl_power -f 24M:1700M:1M -i 10 -c 50% -g 50 -e 40m -F 9 >scan_24_1700.csv

when both scans are done - plot the graphs:

./heatmap.py scan_800_1400.csv scan_800_1400.png
./heatmap.py scan_24_1700.csv scan_24_1700.png
13 Likes

Great! Thanks Tom. Waiting for results.

1 Like

I can’t wait to read and see the results.
You are putting an enormous effort into this.
Thank you in advance.

1 Like

thanx for watching - results are ready within 10 minutes :slight_smile:

although your site already seems to be setup very very well when i look at your range/message graph!

Power spectrums are a simple way to see if you are getting enough attenuation at your particular site but I would not generalize from there to other sites. You’re not really measuring anything new that the individual filter response curves don’t already tell you.

(+ it doesn’t measure how the receiver actually performs, either)

1 Like

that’s why i wrote ‘my results’ and described where my site is located. a real world test never can be generalized and the same applies to every laboratory setup. but it simply proves the filter response lab-measurings. i did test the same setups in one to one mode while running dump without mlat over the last weeks and the numbers there show the same big picture as above graphs.

what the test definitely shows is that the uputronics and jetvision stuff very good performs. the fa-filter curve was not really the best idea one could have - and this applies to more sites worldwide than it does not apply.

Yeah, if you are in an environment where the FA filter is too broad you will need to go with one of the more expensive ones. Some environments are OK with just the FA filter. Some environments don’t need a filter at all (especially anywhere that’s not urbanized). You can’t really generalize, you need to look at the local noise spectrum (in more detail than “really heavy noise”, whatever that means!) to pick an appropriate filter.

1 Like

The bandwidth of the RTL dongle is about 2.4MHz. To capture these images the dongle has to step through a few MHz at a time and measure the power levels at each step. It looks like there is some scaling going on as it goes through from low to high frequencies. There also seems to be some scaling between filters.

A good rule is to use as much filtering as you need. It is hard to know how much filtering you need unless you run one of these power spectrum test.

absolutely that’s why i always recommend to do a noise scan before buying a filter :slight_smile: regarding the fa-filter i think it will fit on overall noisy environments in the usa where gsm960 is not used. but unfortunately big parts of the world use this band that is very near to ads-b. with heavy noise i refer to e.g. very big cities with high power transmitters on all bands and masses of low-power transmitters. what i found most interesting was how unbeliavble good the cavity filter nearly sculpts the 1090mhz out of the whole band and in no way attenuates ads-b signals.

You quoted 0.5dB earlier; what’s the actual value?

Did you measure response curves from your data? (As David points out, you need to be careful with scaling; also make sure you set the receiver gain low enough that it captures the strongest signals cleanly)

So basically a higher noise floor? So long as your filter excludes the higher-power transmitter I don’t see how a tighter filter really helps much with a higher noise floor.

i forgot to specify the gain i used - now added above in description - was 50 in all runs :slight_smile: david when you say ‘A good rule is to use as much filtering as you need’ this was always my credo too. but i have to admit that meanwhile i’d say if one can afford a good cavity-filter like the above then this filter nearly never is a mistake. the 0,5db attenuation is nearly negligible and moreover it adds protection to the receiver.

50 is almost certainly too high for a quantitative comparison of signals, all you can really tell with that much gain is that there are some stronger signals around, not their relative strengths so much.

nope - i don’t have the needed equipment until i get the lime mini :blush: so i simply confide in guenters measurings and skills with his very expensive rohde & schwarz equipment. see below graph …

50 is about the same gain i get most aircrafts and messages when running dump1090 - that’s why i chose this setting for the above tests too. probably this is different to your site that is more urban than my bavarian country site …

1 Like

The two main specs of a filter is how much of the signal you want is let through (passband) and how much of the signal you don’t want are attenuated (stop band).

The FA filter has a pass band attenuation of 2dB and a stop band attenuation of 40dB. This makes it a good filter for ADSB.

The cavity filters are very good but then you are comparing a 4x more expensive filter. I usually see specs of cavity filters in the 0.2dB to 0.7dB in the passband and about 40-100dB in the stop band.

1 Like

yes price-wise you cannot compare those! but what impresses me most is that this very very old and mechanical concept even today beats every single newer technology :slight_smile:

I think you are missing my point. 50 is too high for the GSM signal, not for the ADS-B signal.

I meant that you could compare the relative power in the different power spectrums you collected to approximate that curve. (Which goes back to what I was saying about not really measuring anything new - if you have Guenter’s curve, just multiply your original spectrum by that curve…)

absolutely to high for the gsm-signal! but as i wanted to measure the live situation i have when running dump1090 i chose 50. if i would want to connect or listen to gsm 960 i would of course use a lower gain setting. in addition the first three runs were without use of any kind of amplifier - so i’d say 50 is there not a high value at all …

did you and david already do longer test-setups with the uputronics lnas and guenters cavity-filter?