Finally, progress! (evidence based radio testing method) (and capacitors)

  • Like many, I struggled for a long time to get things working reliably. Finally some progress the last couple days. The biggest leap forward for me was learning about the PingPong sketch, and using it to implement evidence based testing. Along with a couple other small tricks.

    If you are having radio trouble (or reliability trouble in general), I highly recommend starting out with the PingPong sketch, aka GettingStarted.ino example in the RF24 library. In fact, I think I will be flashing this first on all new radio nodes, to test my soldering / assembly, as well as positioning of the node initially, etc.

    As an aside, I also started using PlatformIO recently, I personally find it much less confusing than the official Arduino IDE (and now I can use a real editor! 😁 ). I only mention it because that is the serial monitor command you will see below.

    I am also on GNU/Linux, so I collected my data with a simple tee command on the output of pio device monitor, like so:
    $ platformio device monitor -b 115200 -p /dev/ttyUSB<X> | tee test<Y>

    I then make a copy of the file and call it test<Y>-trimmed, which I then edit by hand to remove header lines at top and also the lines like "*** CHANGING TO TRANSMIT ROLE -- PRESS 'R' TO SWITCH BACK
    ", etc. So we end up with only result lines like the following:

    Now sending 92111...ok...Got response 92111, round-trip delay: 28
    Now sending 93139...ok...Got response 93139, round-trip delay: 29
    Now sending 94170...ok...Got response 94170, round-trip delay: 34
    Now sending 95205...ok...Failed, response timed out.
    Now sending 96408...ok...Got response 96408, round-trip delay: 23
    Now sending 97432...ok...Got response 97432, round-trip delay: 23

    Now that I have a "trimmed" file containing only lines of results, I wrote a small bash script to calculate some things from there a little more easily:

    lines_total=$(cat "$1" | wc -l)
    lines_response=$(grep 'Got response' "$1" | wc -l)
    percent=$(echo "100 * $lines_response / $lines_total" | bc -l)
    artd=$(grep 'round-trip delay: ' "$1" | rev | cut -c -2 | rev | awk '{ sum+=$1 } END { print sum/NR }' )
    printf 'Lines of Response: %s\n' "$lines_response"
    printf 'Lines Total: %s\n' "$lines_total"
    printf 'Response %%: %.2f\n' "$percent"
    printf 'Average Round Trip Delay (ARTD): %.2f\n' "$artd"

    I save it as calculate, make it executable ($ chmod +x calculate) and then execute it with the name of the filename to process like $ ./calculate test1-trimmed which would return something like:

    Lines of Response: 801
    Lines Total: 1016
    Response %: 78.84
    Average Round Trip Delay (ARTD): 25.75

    I then plug these results into a table in Orgmode in Emacs, carefully accounting for each variable I thought might be relevant:

    | TN | SL | RL | STC   | SC | SCL | RTC | RC | RCL |   LR |   LT |    R % |  ARTD | N |
    |  1 | 1  |  1 | 1,3,4 | 1  | 12  | 2   | 1  | 21  |  878 |  906 |  96.91 | 28.16 |   |
    |  2 | "  |  2 | "     | "  | "   | "   | "  | "   |      |      |        |       |   |
    |  3 | "  |  " | "     | 2  | "   | "   | 2  | "   |  801 | 1016 |  78.84 | 25.75 | 1 |
    |  4 | "  |  3 | "     | "  | "   | "   | "  | "   |  910 |  959 |  94.89 | 23.68 |   |
    |  5 | "  |  4 | "     | "  | "   | "   | "  | "   |  853 |  946 |  90.17 | 25.22 |   |
    |  6 | 6  |  " | "     | "  | "   | "   | "  | "   |  871 |  875 |  99.54 | 23.73 | 2 |
    |  7 | "  |  5 | "     | "  | "   | "   | "  | "   |   28 |  289 |   9.69 | 40.39 |   |
    |  8 | "  |  7 | "     | "  | "   | "   | "  | "   |  684 |  684 | 100.00 | 22.91 | 3 |
    |  9 | "  |  " | "     | 1  | "   | "   | 1  | "   |  600 | 1042 |  57.58 | 34.47 |   |
    | 10 | "  |  " | "     | 3  | "   | "   | 3  | "   | 1014 | 1051 |  96.48 | 24.58 |   |
    | 11 |    |    |       |    |     |     |    |     |      |      |        |       |   |
    - Key:
      - Abbreviations:
        - TN = Test Number
        - SL = Sender Location
        - RL = Receiver Location
        - STC = Sender Test Conditions
        - SC = Sending Capacitance
        - SCL = Sending Cable Length (mm)
        - RTC = Receiver Test Conditions
        - RC = Receiver Capacitance
        - RCL = Receiver Cable Length (mm)
        - LR = Lines Response
        - LT = Lines Total
        - R % = Response %
        - ARTD = Average Round Trip Delay
        - N = Notes
      - Keys:
        - Locations:
          1. desk in front of keyboard
          2. coffee table living room
          3. dining room table
          4. kitchen counter by stove
          5. floor by front door (previous door switch radio location)
          6. On top of both CPU towers, then plastic box
          7. front door but higher up
        - Test Conditions:
          1. Arduino Nano 5v
          2. Arduino Pro Mini 3.3v
          3. Mismatched length power cables
          4. ferrite choke on USB power cable to computer
        - Capacitor(s):
          1. single 4.7uF capacitor
          2. 1uF + 10uF capacitors
        - Notes:
          1. A lot of interference from me moving my chair/body(?) in between and banging tools on the
          2. Antenna in vertical plane, face toward dining room table.
          3. Took some effort to align plane of receiving antennae face to face.

    Now, I am no where nearly as talented in electronics and radios as some of the guys on this forum, however I do know how to follow basic scientific principles and do experiments... And by doing so, I have been able to learn quite a lot about radio propagation, electronics (ideal decoupling capacitor values), as well as things particular to my own site conditions here. And this is what I wanted to share. Even if you are not an expert, you can do the same and get a much better result.

    For me, I feel like I am finally making some progress (instead of throwing darts in the dark)!

    A few other little details. I want to talk about decoupling capacitors. I know it is already mentioned frequently, but doing experiments as above you can test and really see the difference for yourself.

    First, how to attach the caps? I saw in someone's YouTube video (Great Scott! I think it was) where he had soldered the cap to the top part of the nRF board, I mean where the stub of the header pokes through the board, where you would be soldering to attach the header, opposite side of where you would attach your dupont cables. I thought this was a great idea and I started doing it, too. Keeps it out of the way, and insures a good electrical connection. I never really liked just jamming the leads of the cap into the back of the dupont connector housing. I never was sure it was making a good connection, the wires are easy to bend and they are different lengths, etc... just a big hassle. This way is much better!


    Secondly, about decoupling capacitor values. I have heard all kinds of stuff thrown around, from 4.7uF to 47uF to 100uF, and then some people even saying to combine a 0.1uF and a 10uF in parallel. I get the impression, that in most of cases (except the parallel case, or in the case of you few EEs out there πŸ˜‰ ) that most people are just guessing, and/or parroting stuff they heard somewhere.

    I did a bit of research on this, even started watching one of EEVBlog videos about it (until I started dozing off, lol). Now, I am no EE for sure, but there appears to be something to the parallel thing, which I was able to bear out in testing. Check some of my results above where I purposefully kept all conditions the same, except for swapping out radios with different decoupling capacitor setups. The results (for me at least) were dramatic. For instance look at tests 8 and 9. Now I didn't have a 0.1uF cap on hand, so I used the smallest one I could find which was 1uF + a 10uF. And with that arrangement, 100% Response %! Compared to 57.58% with a single 4.7uF cap. This is at maybe 20m distance inside, through 1 or 2 walls with metal studs and drywall.

  • I wanted to test my theory that two different value capacitors in parallel was somehow better than one of a similar capacity. So with everything exactly the same as I had left it last night, I soldered up two new radios, except this time with only a single 10uF electrolytic 50v capacitor, instead of 1uF + 10uF.

    Results can be found in "test 10" above. Looks slightly less reliable at 96.48% than test 8 which was 100%!

    Now, I am not going to the trouble of using the exact same radios, desoldering components, etc... I suppose there is a chance that some of these radios might be "better" in some way than others. They did all come from the same batch at least, FWIW...

    Perhaps one of you professional angry pixie wranglers out there can verify that there is in fact something to the "two capacitors of different values in parallel" theory?

    But for me, this is a dramatic improvement over what I was getting before (test 7 being closest approximation), so I think I will get back to building a new gateway and some new nodes, putting to work some of the things I have learned here.

    Cheers! 🍻

  • Mod

    Nice work. I took the liberty to reformat the table into markdown, which is easier to view in the forum.

    1 1 1 1,3,4 1 12 2 1 21 878 906 96.91 28.16
    2 " 2 " " " " " "
    3 " " " 2 " " 2 " 801 1016 78.84 25.75 1
    4 " 3 " " " " " " 910 959 94.89 23.68
    5 " 4 " " " " " " 853 946 90.17 25.22
    6 6 " " " " " " " 871 875 99.54 23.73 2
    7 " 5 " " " " " " 28 289 9.69 40.39
    8 " 7 " " " " " " 684 684 100.00 22.91 3
    9 " " " 1 " " 1 " 600 1042 57.58 34.47
    10 " " " 3 " " 3 " 1014 1051 96.48 24.58

  • It's great to see that you are putting in the effort to thorougly test and optimize your setup. I like that! πŸ‘

    Now, I'm also not an EE myself - in fact, I consider myself as a beginner still, so please correct me if I'm wrong. The large (electrolytic) capacitor not only provides a good portion of the juice when it is quickly needed to keep the voltage level up, e.g. if the radio starts transmitting - it also smoothes low-frequency changes in the voltage, but their high-frequency characteristics are rather poor. To also decouple high-frequency noise, you'd want to add a small (ceramic) capacitor, typically 100nF, aswell. Together they provide a smoother voltage than just a single capacitor would. If the larger of the two is 10, 47 or 100 Β΅F shouldn't matter too much, as long as the power source is able to keep up. Of course, if you go further into the details, there's much more to it.

    I provide all my NRF24 nodes with (at least!) 100Β΅F of bulk capacity via electrolytic capacitors - even if I'm heavily space constrained, like on tiny coin cell based nodes. I also always add a 100nF ceramic as close to the radio as possible and usually also an additional 10Β΅F ceramic right next to it. This setup is working great for me: my "NACK rate" is usually below 0.1%. For example, my outdoor weather station has sent 5072 messages over the last seven days, of which four failed to transmit initially (< 0.08%).

    I monitor the reliability of my nodes by simply checking the return value of send() and sending an error count if the transmission failed (ensuring that this message is successfully sent, or else I retry until it worked). All data is stored in a time-series database, so I can simply query the number of data points in a specific time frame as well as the number of reported errors.

  • @TRS-80 said in Finally, progress! (evidence based radio testing method):

    Perhaps one of you professional angry pixie wranglers out there can verify that there is in fact something to the "two capacitors of different values in parallel" theory?

    Two capacitors in parallel lowers the ISR of the capacitance attached to the circuit, meaning it "reacts faster" to the changes in voltage and acts as a better filter. As BearWithBeard said, you can achieve a similar effect by combining electrolytic and ceramic capacitors. If you have a lot of LF and HF noise, that might give you a better result.

  • Thanks for the great replies, guys!

    @mfalkvidd said in Finally, progress! (evidence based radio testing method):

    I took the liberty to reformat the table into markdown

    I hope via some automated tool, and not manually! πŸ˜• (I know Emacs can export Org tables several different ways). But I will keep in mind for next time, thanks (didn't even occur to me until now).

    @BearWithBeard said in Finally, progress! (evidence based radio testing method):

    It's great to see that you are putting in the effort to thorougly test and optimize your setup. I like that!

    It's not like I had a choice, I could not get anything to work before, and gave up in frustration! πŸ˜„

    @BearWithBeard said in Finally, progress! (evidence based radio testing method):

    my "NACK rate" is usually below 0.1%

    Well done! That is really great man, and I for sure will be implementing something similar with checking the return value of send() and using a time series database.

  • @BearWithBeard said in Finally, progress! (evidence based radio testing method):

    I monitor the reliability of my nodes by simply checking the return value of send() and sending an error count if the transmission failed (ensuring that this message is successfully sent, or else I retry until it worked)

    Would you mind sharing your code for this bit? It would save me a lot of time figuring it out from scratch, as I am still in the "searching the Internet and putting code together like legos" phase. πŸ˜„

  • Well, a much neater solution than what I'm using on my nodes was developed recently in this thread (relevant post with code here), which makes use of MySensors' internal indication system:

    void indication(indication_t ind)
      switch (ind)
        case INDICATION_TX:
        case INDICATION_ERR_TX:
    void loop()
      static unsigned long last_send = 0;
      if (millis() - last_send > REPORT_INTERVAL) {

    It has the benefit, that it also works on repeaters and the GW, because MySensors handles everything under the hood (besides the number incrementation and sending of the values, of course).

    I didn't feel like updating all my working nodes with this yet, so I'm mostly still doing it manually. I also prefer to reset the error counter to zero when the value has been reported. I feel it's easier to keep track, when and how many errors appeared that way. And it's nicer to visualize in a graph than an incremental value.

    On simple nodes, sending only one or two values, I keep it quite basic:

    uint8_t txErrors;
    void loop() 
    	// Increase counter, if send() returns false
    	if (!send(msg.set(value))
    	// Send msgTxErrors only if there have been errors to save battery
    	if (txErrors) 
    		// Only reset counter if error count has been sent successfully
    		if (!send(msgTxErrors.set(txErrors)) 
    			txErrors = 0;

    For larger nodes, like the weather station, I have a wrapper around send() to keep the code tidy which basically looks like this:

    uint8_t txErrors;
    struct Measurement
    	Measurement(const char *n, float t, uint8_t d, bool c) : NAME(n), THRESHOLD(t), DECIMALS(d), SEND_ALWAYS(c){};
    	const char *NAME;
    	float value = 0;
    	float lastValue = -1;
    	uint8_t silentCount = 0;
    	const float THRESHOLD;
    	const uint8_t DECIMALS;
    	const bool SEND_ALWAYS;
    // For each child / sensor 
    Measurement Lux{"Illuminance", LUX_THOLD, LUX_DECIMALS, LUX_SEND_ALWAYS};
    void transmit(const MyMessage &msg, Measurement &m)
    	if (...) // Check if threshold crossed, etc ...
    		if (!send(msg.set(m.value, m.DECIMALS)))
    void loop() 
    	// For each child / sensor 
    	transmit(msgLux, Lux);
    	if (send(msgTxErrors.set(txErrors)) 
    		txErrors = 0;

    If you're using repeaters and are worried that they may loose the node's messages on the way, you could request an echo from the target (GW usually) to make really sure that the message arrived:

    void loop() 
    void receive (const MyMessage &msg) 
    	if (msg.isEcho()) 
    		// Msg (and valid echo) arrived

  • Not using repeaters, but I am aware that the return from send() is for next hop only. In fact that was part of what convinced me to skip the repeater function in the node I am currently working on, even though it is centrally located and always powered. Well, that plus my radio testing indicating a repeater was not needed as long as I keep all the radios high up on the walls.

    Anyway, I gave it a quick glance, looks nice! I think I might agree with you about the method of logging the time series data, but need to think about it still. I'll dig into it (and that other thread) later on as I'm knee deep in something else at the moment. πŸ™‚

    Thanks for posting up your code! I've pulled enough hair out by now, anything to ease the pain is always greatly appreciated! 🍻

  • In the meantime I discovered a much more portable, easier to use, and even more feature filled implementation of essentially the same ideas as my OP in nRF24Doctor.

    Of course, doing the way I did in OP requires no special equipment really, but if you get into MySensors enough, it's probably handy to have a dedicated hardware device for such common issues. I am now quite sure I will build one, sooner or later, as they are only few bucks worth of quite common parts. Worth it, to me!

    nRF24Doctor was apparently put on GitHub, including, PDFs of the PCBs. I thought it would be handy to have KiCAD / Gerber files of them instead, so I asked them about that (or even putting it on for that matter), I guess we will see what they say!

  • Nice video here about different types of capacitors, and why you see .1uF used so often in digital circuits:

    Why electrolytic capacitors are actually kinda crappy πŸ’© – 07:21
    β€” Afrotechmods

    Credit to @NeverDie, who originally posted it here.

  • Hero Member

    @TRS-80 There's an argument for putting two different value capacitors near the nRF24L01 (or whichever radio you're using). A 0.1uF bypass cap to help filter out noise, and a higher value cap as insurance against possible battery ESR and/or inductance from possible long battery leads/traces. For example:

Log in to reply

Suggested Topics