easy PCB dipoles for 433, 868 and 915MHz

[captcha]

This weekend I finally had access to a VNA to tune my PCB dipoles for 433, 868 and 915MHz.

The 433 and 868 antennas were attached to the VNA with a 22.5cm length of RG174. On 433MHz this is a half-wavelength (VF=0.65) and two half-wavelengths on 868MHz. For 915MHz I only trimmed the feedline an extra 5mm to 22cm which makes it close to two half-wavelengths.

During testing, the antennas simply were left hanging down, away from any nearby metallic or organic objects.

The following graphs are the resulting plots for each antenna:



433MHz
----------
The above plot is for the dipole tuned for 433MHz. The blue line is SWR and shows a dip around 434.8MHz. Close enough for using it on 433MHz. The SWR (top left, in blue) shows a value of 1.11:1 for this frequency. The red line shows an impedance at this frequency of 45.62 Ohms, reasonably close to matching it with the radio which expects 50 ohms. The bandwidth of the antenna with an SWR of 1.5:1 or less is about 15MHz (from 427 to 442MHz). More than sufficient to cater for any variations in operating frequency.

- fed with one half wavelength (22.5cm) RG-174
- bandwith below 1.5:1 SWR = 15MHz
- each dipole leg is 13.3cm






868MHz
----------
The above plot is for the dipole tuned for 868MHz. The blue line shows an SWR of 1.1:1 at 867MHz. The dip is much wider than on 433MHz and has a bandwidth of about 51MHz (838 to 889MHz) for an SWR of 1.5:1 or less. The red line shows an impedance of 47.48 Ohms for this frequency, so a fairly close match for the radio.

- fed with one wavelength (22.5cm) RG-174
- bandwith below 1.5:1 SWR = 51MHz
- each dipole leg is 6.4cm






915MHz
----------
The last plot is for 915MHz and shows an SWR of 1.29:1 for this frequency. The lowest SWR is slightly lower in frequency, but negligible for our purposes. I just left the marker at 915MHz. The antenna has a bandwidth of 43MHz (888 - 931MHz) for an SWR of 1.5:1 or less. The red line shows an impedance of 42.64 Ohms at 915MHz.

- fed with one wavelength (22cm) RG-174
- bandwith below 1.5:1 SWR = 43MHz
- each dipole leg is 6.0cm



The tests were performed with the SDR-Kits DG8SAQ Vector Network Analyzer 3E (site: https://www.sdr-kits.net/)


(Prototype PCB dipole attached to the VNA)

Now that I have accurate measurements of how long each leg needs to be I have updated my PCB designs and uploaded them to OSH Park for the whole world to grab and reproduce: https://oshpark.com/profiles/captcha. Just order the ones you need and OSH Park will take care of the rest.

On the 868MHz board I have also added cut marks for trimming the antenna for 915MHz use. The 433MHz board has cut marks for both 915 and 868MHz. If you are going to trim the tracks, make sure you remove the excess outside traces.

Based on the test results, I calculated the velocity factor of the FR4 for a single-layer printed dipole (track width 1mm, one once copper) to be about 0.77.

All these designs have solder pads for an edge-style SMA connector:

(Prototype board with SMA edge connector)

I ordered the SMA connector from RS with the part number 526-5785 (http://au.rs-online.com/web/p/sma-connectors/5265785/)

In the PCB designs I have deliberately left a solder mask over of the bottom sma connector ground pad. I did this purely to minimise corrosion in case you wish to solder leads to the pads instead of using the edge SMA connector. The solder mask is fairly easy to scrape off.

Some general things to keep in mind about these designs:

 - This is a half-wave dipole and therefore it does not have a 50 ohm feedpoint impedance. You will have some rf coming back to the radio on transmit but due to the rigid and balanced PCB based design and the accurate calibration of the length of the dipole legs using the VNA this will be minimal.

 - To keep the impedance mismatch from getting worse it is important to only use feedline lengths of (multiples of) a half wavelength (compensated for the appropriate velocity factor) for the frequency of interest. See the picture in this post to see the effects of different feedline lengths: https://lowpowerlab.com/forum/rf-range-antennas-rfm69-library/small-loop-antennas-433-mhz/msg14061/#msg14061

 - coax feedline losses are siginificant at these frequencies, use the lowest number of half-wave multiple lengths possible.

 - I tested these dipoles on 1.6mm thick FR4 substrate PCBs with 1 ounce copper from OSH Park. Feel free to use these designs to manufacture elsewhere but it is not guaranteed that the results will be the same. Heck, even OSH Park could end up supplying you with different materials, but this is what worked for me.

I mainly made these designs for myself so that I could get easy access to a low-cost and good performing omnidirectional antenna for 433MHz. I did the 868 and 915 designs as a bonus for the community as I already had everything set up and it only took me an extra half hour or so to come up with the final designs. I'm hope some of you will appreciate it.

[WhiteHare]

Nice!  Thanks for putting your work on OSH Park.  Is there any downside if I were to solder a Moteino directly to your antenna PCB without using a feedline?

[captcha]

Is there any downside if I were to solder a Moteino directly to your antenna PCB without using a feedline?

As usual: 'it depends'.

Although hard to visualize, the common belief is that maximum radiation happens where current is maximum. For a half-wave dipole this current maximum is at the feedpoint (i.e. in the middle of the antenna, where one leg takes a left turn and the other takes a right).

In a quarter wave vertical antenna (what some call the 'monopole'), current maximum is also at the feedpoint. Think of a quarter wave vertical as a dipole but with one of its legs turned into a plane and perpendicular to the other leg. In terms of using it on a moteino, the 'plane' is provided by the GND tracks of the circuit board and the 'monopole' going straight up is just the other leg.

Many people here on this forum (and elsewhere using similar small radio modules) use quarter wave vertical antennas with great success, even though maximum radiation happens in very close proximity to the microcontroller on the pcb. From this point of view I think you should be fine with attaching the dipole antenna directly to the antenna port.

The close proximity of the pcb to the dipole does screw up some of the antenna's balanced characteristic. A balance I specifically wanted to create using a rigid pcb design. This distorting effect would be even more pronounced on higher frequencies (868/915) due to the moteino being a much bigger size relative to the size of the antenna.

[Felix]

@captcha nice work, thanks for sharing the research!

[captcha]

Thanks, happy to share.

I just noticed that I forgot to include the Diptrace design files. Please find attached the files for all three antennas for those who want to have a closer look.

[joelucid]

Very cool!

[raggedyanne]

Nice work 

The right angle traces from the sma need to be ellipse/arc not square 90 , or at least 45 degree.
What is the trace width ?

Trace Curvature Radius >= 3 × Trace Width

[captcha]

Hi raggedyanne,

The copper antenna element traces have a width of 1mm and a thickness of 1.37 mils (at 1 once per square foot). 1.37 thousands of an inch translates to 34.79um.

Excellent suggestion on the bend radius! Never thought of the importance of it, but I'll be sure to do some further research into the matter and incorporate it into the next design.

Great to have you on the forum!

[raggedyanne]

It like running through a maze without the ability of slowing down 99% of the time you hit the wall , but with curved walls you can slide on through without bouncing around 

[Felix]

Right, imagine the noise if the water pipes had 90° angled turns in your house

[joelucid]

Captcha et al,

I'm about to use these antennas as reference in a project. So quick question:

How much directionality will the asymmetries of this antenna cause (dielectric pcb material only on one side, solder mask on the other etc)? Anybody have a sense for that? Captcha maybe you have experience from your use of the antenna?

Joe

[captcha]

Read up a lot since the recommendation to not use 90 degree angled corners in PCB designs.

Check the links below for further reading pleasure.

In short, some of the drawbacks of using 90 degree corners in pcb tracks:
- during manufacturing, they can trap solution in the corners and over-etch, making corners thinner than expected.
- fewer 'fold areas' which can lead to track breaking due to warping stresses (this is mainly applicable to thin, flexible PCBs)
- less reflection at turns for signal propagation in very high frequency applications.
- very inefficient as they often take up more space to route around components

Note that vias (to connect pcb layers) are in fact 90 degree corners and should be avoided in certain RF conditions.



A bend radius of (at least) 3x the trace width is a minimum for very high frequency applications.



For maintaining a 90 degree corner in a very high frequency signal path, a so-called mitre-bend is the superior technique to change direction on a PCB.


From what I've read (the page at the bottom link is most clear on this), the signal reflection effects become an issue at frequencies above 5 Gigahertz. I'm not convinced that these PCB antennas require rf-friendly angles but a case could be made for minimising physical stresses on the tracks. I will update the designs and post the relevant links here.

How much directionality will the asymmetries of this antenna cause

I don't think it will matter much but because I have the antennas already I expect to run a test this weekend (weather permitting) and post my findings here.

Links:
http://www.pcb3d.com/uploads/Flex_Design_Guidelines.pdf
https://www.maximintegrated.com/en/app-notes/index.mvp/id/5100
http://www.ti.com/lit/an/slla311/slla311.pdf
http://www.microwaves101.com/encyclopedias/mitered-bends

[captcha]

Hi Joe,

Captcha maybe you have experience from your use of the antenna?

I have only used them at close range, but your question intrigued me so I got curious and decided to find out for myself. Today was a beautiful day for testing (28 degrees C, 82F) so after uploading some code to two Moteino's and cable-tying the PCB dipole to a wooden dowel I was ready to do some testing.

I just came back from a little walk around my neighbourhood to find out if the PCB causes the dipole to have any kind of directionality.

Short answer: No perceivable directionality.

The base station consisted of a Moteino sending out three encrypted packets at 2-second intervals, with a payload of some telemetry (temperature, voltage, packet count, RSSI). The antenna used on the base station was a quarter wave vertical with 4 ground radials and a vertical whip antenna for 2m (also resonant on 70cm, so no big worries there).

I walked around with another Moteino that had the PCB dipole for 433MHz connected to it with a 22.5cm length of RG-174. This Moteino was battery powered and I taped everything to a wooden dowel with the dipole at one end. This Moteino had a 128x64 oled screen connected to it and gave me instant feedback as to how many packets were received and what the received signal strength was.


The mobile station.

I walked about 1.4km to a point where the signal often got an RSSI between 99 and 103 and reception was intermittent at best. There was absolutely no line of sight between the two stations as many street signs, a big building, trees, metal fences, street lights, overhead power lines and cars were nearby and around (remember, line of sight does not work like a laser; Google: Fresnel Zone for laughs).

Making sure to keep the antenna vertically polarised at all times I alternated between pointing the copper side of the antenna and the pcb (under) side toward the base station 1400m away. I noticed that the signal dropped away completely when there were a lot of cars in between the stations but sometimes the signal would come good again and I could decode three consecutive packets and received the telemetry data fully intact.

At the fringe of reception in this practical experiment I could not see any significant difference between pointing the antenna with the copper side or with the pcb side to the base station. Often I would get decodes with an RSSI of 98, then one with 102, then 99.. It would definitely jump around but I could not see that one orientation would be better than the other.

What I did notice, is that I used this same walk to test quarter wave verticals at both ends and never got this far.

Fun experiment, learned a lot and I got some exercise too! :-)

[joelucid]

Thanks that is definitely good news! I have some ordered which I will try probably next week to measure some difficult environments. I really like the reproducibility of these antennas. Much better than custom made dipoles for reference testing.

Thank you so much for putting them together!

I have one further question for you: you recommended l/2 multiples for the cable length for understandable reasons. I take it you measure the length from the point where the antenna traces turn away perpendicular from the dipole proper, not from the connector, correct?

And yeah - range testing is one of the best activities to stay in shape 

Joe

[captcha]

Interesting question; half-wavelength is between where and where?

from the point where the antenna traces turn away perpendicular from the dipole proper, not from the connector, correct?

If an antenna is not properly matched, then yes, ideally from transmitter to antenna (the part where the traces separate to make up the elements) should be (multiples of) a half wavelength. There is a slight mismatch with these dipole antennas and you want to be as close to the half wave(s) as possible, purely to minimise rf coming back. Remember, if you were to use an antenna with 50 ohms feedpoint impedance, then the length of (50 ohm characteristic impedance) transmission line would not matter (you still want to keep it as short as possible of course).

In my testing I didn't use connectors but soldered the RG-174 directly onto the pcb. With connectors I would probably take off an extra 2 cm from the coax to cater for the length of the connector and pcb traces towards the antenna, whilst making sure the electrical length between transmitter and antenna is a (multiple of a) half wavelength. It may also depend a bit on the type of connector you're using in case you're crimping them yourselves.


This is how far the SMA connector screws on. Reference scale is in centimeters.

If you're planning on using the SMA connectors, eBay sells RG-174 and RG-316 coax with pre-fitted SMA connectors for next to nothing. A common name for these is 'pigtails'. Velocity factor and attenuation are similar, but the biggest benefit of 316 is that the dielectric is Teflon (PTFE) and keeps shape much better when soldering onto it.

[TomWS]

@captcha, good experiment for the dipole, thanks for posting!  One question: did you leave AGC on or have the gain fixed at the RX end?

Tom

[joelucid]

did you leave AGC on or have the gain fixed at the RX end?

In theory it shouldn't matter since the RSSI detection circuit is supposed to work independent of selected gain. In practice it almost never matters since RFM69 lib selects the RSSI threshold so low that RSSI detection triggers immediately when the radio goes into RX just on background noise. It then always selects maximum gain.

If your noise floor is below -110 dBm or you use a RSSI threshold higher than the floor then AGC does select different gains and the supposedly gain independent RSSI measurement apparatus of the RFM69 chip would be put to the test.

Joe

[joelucid]

If an antenna is not properly matched, then yes, ideally from transmitter to antenna (the part where the traces separate to make up the elements) should be (multiples of) a half wavelength. There is a slight mismatch with these dipole antennas and you want to be as close to the half wave(s) as possible, purely to minimise rf coming back. Remember, if you were to use an antenna with 50 ohms feedpoint impedance, then the length of (50 ohm characteristic impedance) transmission line would not matter (you still want to keep it as short as possible of course).

I guess it really depends on where the impedance mismatch is. If the pcb feedline from antenna to connector has a characteristic impedance of 50 Ohm then it should be looked at as extension of the coax and be added to the length of the coax. On the other hand if that short feedline has the same characteristic impedance as the dipole you would only count the coax.

Did you design the feedline with 50 Ohm? If not then likely both connections reflect and it's not so clear anymore what to do.

I could just hook the antenna up to the vna including coax and find the best length that way.

[captcha]

@TomWS

did you leave AGC on or have the gain fixed at the RX end?

To be honest, I haven't played around with any of the settings. I would like to lower the bandwidth and see what range improvements I get but all my results have been based on the default RFM69 library settings, so if I read the library correctly, the AGC is always on.

Oh, not sure if I mentioned it, but I have been using the Moteino R4 with the RM69HW module at maximum power output (100mW/20dBm).


@joelucid

I guess it really depends on where the impedance mismatch is.

Correct, hence my recommendation to use coax at half-wavelength multiples to not have the length of the coax mess things further up. I'm by no means an expert but for those unfamiliar with rf it is the easiesst path to a decent working antenna system.

For those interested, I used the following method to arrive at the lengths for the antenna elements:

- calibrated the vna with the recommended 50 ohm termination resistor
- connected a 25cm piece of RG-174 with an SMA connector, the other end of the cable was just an open circuit
- cut off bits of coax until the plot showed an electrical half wavelength for 433MHz (this was around 22.5cm)
- stripped about 1cm of insulation from the end of the coax and soldered it to the PCB dipole
- I then ran the vna plot and trimmed each leg by the same amount until the lowest SWR for 433MHz was reached
- since 868MHz is practically double the frequency of 433MHz I left the setup exactly as is but now started trimming the dipole legs until I reached the lowest SWR on 868MHz. Of course, the feedline is now two half-waves but for testing SWR this is negligible.
- once the lowest SWR for 868 was found I de-soldered the coax from the antenna and trimmed it about 5mm and re-soldered it back onto the antenna again.
- again using the vna plots I shortened the elements until I had lowest SWR on 915MHz.

 

Did you design the feedline with 50 Ohm?

You mean the antenna? Yes, depending on the frequency I used either one or two half-wavelengths of coax. The mismatch is therefore at the antenna feedpoint.

I understand this is not optimal, but I've heard people say that 'every antenna is a compromise'. For better matching and rf isolation a 1:1 balun would be a logical option. But this would further complicate the design for newcomers and I'm not too sure whether there is actually a problem that it addresses. Adding a balun would also introduce additional losses and I would not have the equipment to properly evaluate such a design. To honour common sense dipole designs I think it's preferred to use a feedline of at least a half wavelength away and perpendicular to the antenna. This is also why I tested the dipoles with feedline attached; simulating a realistic environment.

Hope this helps.

[TomWS]

@TomWS

did you leave AGC on or have the gain fixed at the RX end?

To be honest, I haven't played around with any of the settings. I would like to lower the bandwidth and see what range improvements I get but all my results have been based on the default RFM69 library settings, so if I read the library correctly, the AGC is always on.

Oh, not sure if I mentioned it, but I have been using the Moteino R4 with the RM69HW module at maximum power output (100mW/20dBm).

I asked because of the seemingly random successful vs unsuccessful packets at the extreme of the range.  If the AGC is turned off with fixed max gain, I'd expect a more stable (and measurable) result. You're not bucking against some auto correction algorithm.  As you have usefully observed, environmental conditions like car traffic will influence the results.  Again, good work.

Tom

[captcha]

Thanks Tom,

I wasn't sure what to expect at the edge of reception but the ongoing car traffic seemed to be the most varying factor. I guess I could do another test over a nearby lake with pretty much an unobstructed signal path, which should give me a few km's of range. That won't happen anytime soon but I am curious about maximum range with these PCB dipoles at either end.

In any case, thanks all for your feedback and suggestions.

[TomWS]

I guess I could do another test over a nearby lake with pretty much an unobstructed signal path, which should give me a few km's of range.

I would think the lake would act as an RF reflector of sorts and might even increase the range vs dry land.  Not sure about this, but that comparison would be interesting (although not necessarily useful for most people).

Thanks for your efforts,
Tom

[joelucid]

I would think the lake would act as an RF reflector of sorts and might even increase the range vs dry land.

I have a lake thermometer installed on the distant side of a 150m lake with the gateway in a shed behind the weekend house. I get <3% error rates with an RFM69HW sending at 9 dBm and the house between thermometer and gateway. So I can confirm very good reception on the water in line with your theory. Thermometer is a Moteino with wire, gateway an esp8266 based gateway with custom dipole.

RSSI is around -85 now. I initially ran at 2dBm but that led to lost packets when big ships crossed ...

[joelucid]

Did you design the feedline with 50 Ohm?

You mean the antenna? Yes, depending on the frequency I used either one or two half-wavelengths of coax. The mismatch is therefore at the antenna feedpoint.

I meant the short PCB trace between the two elements of the dipole and the connector. Unless that's a 50 Ohm micro strip you'll get reflections at the SMA connector, too. And if it's not 50 Ohm and not the impedance of the dipole you'd get a first reflection at the SMA connector and then another one at the antenna.

See http://www.analog.com/media/en/training-seminars/tutorials/MT-094.pdf on characteristic impedance of micro strips.

I think I'll just measure out the coax length by trial and error for the first run.

Joe

[captcha]

I got the dimensions for a 50 ohm impedance trace somewhere off the net and believed at the time that they were right (1mm track with, 1.15mm spacing between centre trace edges and GND pad edges). I haven't done much further research into that part of the connection but it fortunately is only a very short section.

Embarrassingly, I'm not able to find the document anywhere and all I can find is something called a 'coplanar strip with ground'. The calculations for the latter don't match up so I'm not sure if I'm on the wrong track there.. (pun not intended)

The values for RealZ on the 433, 868 and 915MHz dipoles (see graphs on first page of this post) are 45.62, 47.48 and 42.64 ohms respectively and these were measured 1x or 2x half-wavelenghts away from the antenna, so it can't be all that bad.

[perky]

To get 50R microstrip trace on a two sided PCB of standard 1.6mm thickness the track needs to be about 3mm wide. It's better I think not to have any ground plane under them and treat those short traces as an extension of the dipole arms, which is possibly what really actually happens.
Mark.

[raggedyanne]

The pads on the board can have as resistative or capacitance effect.

RF / Microwave PC Board Design and Layout
https://www.jlab.org/accel/eecad/pdf/050rfdesign.pdf

Basics in RF and EMF screening and shielding
http://incompliancemag.com/article/the-basic-principles-of-shielding/

[john4444]

Raggedy,
Very good references.
Thanks

[joelucid]

I received my dipoles yesterday and started testing a bit. I got both 433 and 868 Mhz ones. Both antennas are a tiny bit too small, the resonance of the 433 Mhz one is at ~443 Mhz and the 868 Mhz one around 875 Mhz. They both measure pretty much on spot at 433 and 868 if I hold the dipole by its feedline in my hand during measurement. 

The bandwidth of the 868 Mhz antenna is wide enough that it is certainly workable and I did first tests with it.

Now the impedance came out even lower than what captcha measured. I got 39 Ohm for the 868 Mhz antenna. That's probably still ok matching wise, but I wonder why it is so much off the ~75 Ohm a dipole should have. Is it the dielectric? Is it the outer coax that's part of the antenna since there's no balun? Anybody have a theory?

Joe

[captcha]

Hi Joe,

Thanks for the comments and the thorough testing you did, and what a great starting point to further test and improve these designs! It's amazing to see how well these antennas are already performing, given the vast number of variables and designing these things at such little cost.

Both antennas are a tiny bit too small, the resonance of the 433 Mhz one is at ~443 Mhz and the 868 Mhz one around 875 Mhz.

I can think of a few things that could add to a probable cause, but without further testing, have nothing substantial yet to back this up.

Some pointers (in no particular order):

* fault in the gerber files
* different production materials at the PCB manufacturer
* different measurement equipment
* different measurement techniques/environment
* something else..?

I have also just received my batch of antennas this week (only the 433MHz ones) and should be ideal for re-confirming (or not) what you see. Thus far I have only held them up next to my first prototype antenna and they seem to be within 1mm of accuracy of each other. A rough calculation shows me that a frequency of 10Mhz higher (to 443MHz) should be the result of a difference of about 2.3mm shorter on each leg! That definitely isn't the case with the new antennas I just received so something else must be going on.

on spot at 433 and 868 if I hold the dipole by its feedline in my hand during measurement

Interesting.

You probably already know that this is not how I tested these antennas with the VNA (my testing procedure is described a few posts earlier) but it does show that environmental changes can easily affect the performance.

I'm open to any other interpretations others may offer as it seems there are some other variables that have come into play.




Then, things impedance..

why it is so much off the ~75 Ohm a dipole should have.

Some pointers:

* The effects of ground are quite substantial on a dipole.
* Add (almost lack of) thickness of the antenna elements to the equation as well.
* Then there's the feedline that can make all the difference. (is it really 50 ohms characteristic impedance?)

Obvious question, but I have to ask since it's not specifically mentioned: you did tune the feedline in each scenario for electrical half wave(s)?

Mind you, my setup could have been just as prone to all sorts of misinformation and variances in equipment etc.. so I'm definitely not saying who's right and who's wrong. However, with all the information presented I think we're getting very close to coming up with a low-cost, simple and effective design.



In short, I think the resonance is the thing to chase, I wouldn't worry too much about impedance.

I will hook up my new 433MHz antennas to the VNA and post here what it reveals. It could very well be that changes in the design are required. In the next version I also want to: 1) include the modification to change the right angles into smooth bends to avoid reflections and 2) re-do the tiny pcb transmission line (from SMA connector to antenna elements) so it's closer to a 50 ohms characteristic impedance.

[john4444]

Hi Joe,

It is not really surprising that you are seeing somewhat different values for the antennas.
There are a lot of variables affecting antennas that are difficult to control.
The environmental factors are especially difficult.

Just to complicate a difficult subject even more, sometimes, perfect impedance-matching
in the transmitter antenna does not always result in the best received signal level in the receiver.

It is very difficult to quantitatively evaluate different antennas.
I would urge you to compare (qualitatively) the antennas by the received signal strength level
(RSSI) or maybe better by the number of missed packets per 100 or 1,000 total packets.
It will probably be necessary to operate at low transmit power levels to get reasonable
numbers of missed packets but that shouldn't be much of a problem with RFM69's.

Maybe perfect impedance-matching will result in the best RSSI or fewest missed packets,
but it is more likely that some other factor is more significant.

Captcha,
Luckily, I got a notice of your comment as I was responding to Joe.
Nice job on the antennas.
I am speculating but, I believe that you are probably 95% as good as you can get.
Any significant improvements are unlikely.
However, most users will be able to compare for themselves the antenna performance
as I outlined above without any specialized equipment.
Keep up the good work.

Good Luck, John

[captcha]

Thanks all for your feedback so far.

@ raggedyanne

oscillator is of low quality

Of the VNAs? I very much doubt that. These devices are made specifically to be spot-on when it comes to frequency measurements. Remember, these tests are done without the RFMxxx board attached, it's purely the antenna system's characteristics we're measuring.


@john4444

I believe that you are probably 95% as good as you can get.

Ooh.. I like the sound of that! :-)

The original idea was to come up with a design for a low-cost, well-performing antenna within the constraints of a) my access to measuring equipment, b) catering for many different end-user environment scenarios and c) a super-simple design that could be easily and accurately replicated (thanks to cheap pcb manufacturers like OSH Park).

I like to think that version v1.0 of these antennas is already a huge improvement over the 'monopole' wire that has been suggested in other forum posts. Yes, this is a design that requires a bit more care to set up properly but I believe it is a step in the right direction to properly satisfy the radio's swr and impedance needs. Once you have that you're almost guaranteed to have good antenna.

Thanks for the extra eyes looking at these designs, I will try to stay in touch to find out what's going on and why Joe's measurements were slightly off.

[captcha]

Joe mentioned that when he hooked up these dipoles to his VNA (Vector Network Analyzer) he noticed that they were resonant at frequencies slightly above the desired frequencies. He did not obtain these results using the radio modules whatsoever.

Hope this clears things up.

[WhiteHare]

Now that I have accurate measurements of how long each leg needs to be I have updated my PCB designs and uploaded them to OSH Park for the whole world to grab and reproduce: https://oshpark.com/profiles/captcha. Just order the ones you need and OSH Park will take care of the rest.

I received OSH Park antennas (thanks for the link).  You kindly gave the link for the SMA connectors ("I ordered the SMA connector from RS with the part number 526-5785 (http://au.rs-online.com/web/p/sma-connectors/5265785/)"). 

So, all well and good.  The only thing I need now is the right cable, with connectors, of the correct half-wavelength.  Is there a source where those are already made-up and ready to buy as well?  I could probably throw something together, but something that's professionally made would presumably have less insertion loss at the connector junctions, and I'm guessing the cost wouldn't be much anyway.  Or am I better off skipping the connectors and just soldering the cable at both ends (one end being the antenna and the other end being the Moteino)?  Would that have less insertion loss, or more?  It does have the virtue of being simple: all I would need is some 50-ohm coax of the correct length, is that right?

[captcha]

I received OSH Park antennas

Awesome! 

cable, with connectors, of the correct half-wavelength.  Is there a source where those are already made-up and ready to buy as well?

I haven't seen them as pre-cut for ISM-band frequencies, but in my projects I have been using the RG-174 (or RG-316) leads with ready-made SMA connectors at either end as found on eBay. I then cut one side off to length and solder that to either the radio or the antenna. I simply haven't had the need yet to have proper connector-terminated radios AND antennas in one application.

all I would need is some 50-ohm coax of the correct length, is that right?

Correct. Velocity factor for a lot of coax is around 0.66 so that should be all you need to determine how long the feedline needs to be.

As an example for 433MHz:
- Full wave in free space: 300 / 433 = 0.692 m (69.2 cm)
- Half wave: 69.2 / 2 = 34.6 cm
- Compensated for 0.66 velocity factor of RG-174 coax: 34.6 x 0.66 = 22.9 cm

If you look at the first post on the first page of this post you'll see that I ended up with 22.5cm for a half-wave on 433MHz. There I had the benefit of an antenna analyzer to determine more precisely the velocity factor, but in this example 22.9 cm is good enough if you stay below a couple of half-wave multiples.

Soldering to either the Moteino or PCB Dipole should be fine. Just be aware that RG-174 dielectric is not as heat-resistant as RG-316. I wouldn't worry too much about connector insertion loss, but soldering tends to corrode less between connected surfaces. If you are going to place the project outdoors I would investigate sealing compounds but that's a whole nuther story.

[WhiteHare]

So, I don't need to subtract the velocity adjusted length of the PCB trace from the velocity adjusted 1/2 wavelength of coax in order to get the proper length of coax?  This is all foreign territory to me, but I had started to imagine maybe they acted together as a single antenna.  The layman's analogy I was working from to try to understand this was: if the entire dipole antenna were built on a PCB, but you constructed a Faraday cage around the middle of it (which would represent the coax part), such that the parts that stick out of the Faraday cage are equal in length to the traces on the PCB you designed.  i.e. from a resonance standpoint, it's as though you built a complete dipole antenna on the PCB, but you're only allowing the protruding ends to be stimulated by the RF in the environment.  Whereas what you're constructing would be that, except the Faraday cage engulfs the entire imaginary 1/2 wavelength PCB antenna (with length adjusted for the velocity of the PCB medium)--which obviously wouldn't work at all as an antenna--so then the length of the PCB antenna traces are extended so as to protrude the all-engulfing Faraday cage by the length of the antenna traces you have on the actual PCB.  However, from a resonance point of view, doesn't that make each leg of the antenna too long?  i.e. don't you want both the parts of the trace (the part inside the Faraday cage and the part extending beyond it) to sum to the velocity adjusted 1/2 wavelength?

I should probably just shut-up and follow the directions, but it seemed like a fair analogy, so I'm just wondering if it has any merit to it, or if I'm (more likely) completely misunderstanding the theory that's driving the design.

[perky]

If the feed point for the dipole was perfectly matched, and the output impedance of the transmitter was perfectly matched, and the coax was ideal, then it shouldn't matter how long the coax is. So I think the 1/2 wavelength criteria is about minimising the effects of reflections back from the feedpoint and those re-reflected back from the transmitter. How important that really is compared to other things though is debatable, I don't personally think it matters that much especially if the transmitter output impedance is well matched.

Mark.

[WhiteHare]

So I think the 1/2 wavelength criteria is about minimising the effects of reflections back from the feedpoint and those re-reflected back from the transmitter.

Thanks!  I'll run with that as the explanation.

[ChemE]

Might rev2 of this design be available anytime soon?  I'd love to do a 915MHz design myself but my free version of Eagle limits me to a maximum board size of 100mm and I need 120mm for the traces alone.  Also, the boards could be shrunk to 0.25" wide and drop the price by half unless 0.5" is used because of shipping concerns.  Finally, from reading through the thread again it seems the dipole length and feed may benefit from slight tweaking.  Thanks for the work done to this point though!

[captcha]

Certainly.

The boards as shared on OSH Park are all 12.2mm wide and this was done to cater for the placement of an SMA connector. With that connector I don't think I could shave off more than perhaps 2mm on the width.

Of course, I could forgo the SMA pads entirely and just have small soldering tabs (probably through-hole for a bit more strength). That would allow me to make them less wide and nullifies the suggested improvements of square corners and 50 Ohm 'stripline' track impedance.

Any suggestions for size, quantity and placement of mounting holes? Perhaps two holes next to the coax for a small cable-tie?

Once I have that I will order a trial run with slightly longer tracks so that I can trim them for lowest swr on 915MHz and then publish the boards for all to grab as per usual.

[ChemE]

...
Of course, I could forgo the SMA pads entirely and just have small soldering tabs (probably through-hole for a bit more strength). That would allow me to make them less wide and nullifies the suggested improvements of square corners and 50 Ohm 'stripline' track impedance.

Any suggestions for size, quantity and placement of mounting holes? Perhaps two holes next to the coax for a small cable-tie?

Doing away with the SMA pads in favor of a few vias to solder the coax to sounds great to me.  Fewer parts to buy and less chance of corrosion long term.  Obviously that gives up easy disconnection but I don't see that as much of a feature.  You've already been thinking about this, your suggestions are spot on.  For me, I'd like a board closer to 120mm x 6.35mm wide which is OSH Park's minimum dimension simply to push cost down to the floor.  I know 120mm isn't possible but that board would be $5.91 for 3 shipped which is absolute peanuts.  How small can you go?

[perky]

I would consider using U.FL connectors, you can buy very cheap U.FL coax 'patch' cables for a few dollars (with or without a second U.FL plug on one end). Admittedly this would be for relatively short cables (a few 10s of cm) because those cables have losses, but it would allow putting the antenna in a box very easily.

Mark.

[captcha]

U.FL.., yep, can do that as well. (hey, it rhymes).

I haven't ventured into using those before, but I've seen them inside laptops for the WiFi and Bluetooth feedlines. Yes, they are really nice and small, it's just that I'm not overly confident they will still 'snap' on after 20x disconnects. But for small, neat and short runs of connections they would be awesome. I'll add them to the designs and share them all (SMA, through-hole and U.FL).

Anything else I could add before I send them off for production? This is the time.. ;-)

[ChemE]

Maybe you are already planning on doing so, but I recently used SparkFun's SOIC-8 BoB and really liked that the long pads had vias placed in the center to help them soak up the solder instead of blobbing out to hit the next pad and make a junction.  I'm not accustomed to fine-pitch soldering (like the U.FL connector being discussed) and felt that this design element helped my clumsy hands do the job.

I'll save you the click...


Other than that, the only other thing I could suggest is just to keep the boundary as small as is possible so the cost is minimized since this is meant to be an easy reference design for the masses if you will.

[perky]

Well, not relating to this but if I had the time I'd really like to have a go at laying out one of these cross meandered dipoles, with and without a U.FL patch cable to introduce a 90 degree phase shift between them to make it circularly polarized, as it may well be significantly more omnidirectional than a single dipole. Gain would be less of course

http://cearl.ee.psu.edu/projects/assets/Project3/Project3_3/Stochastic%20Antennas1.pdf

Getting this to simulate in NEC would be great. That's if you've run out of things to do

Mark.

[joelucid]

My results with your dipole and a long (l/2) coax to the board have been great. But as soon as I mount the antenna close to the espgw board I get significant noise. I've been working on a new version of my gateway to address these issues. Two of the changes I've made might be interesting to you:

1.) I decided to use a balun with the dipole to try and get better noise immunity. I settled on the HHM1522A7 (https://product.tdk.com/info/en/documents/data_sheet/rf_balun_hhm1522a7_en.pdf) for 868 mhz.  The reason I think it might help is that without it the outer surface of the coax and the ground plane of the board acts as third leg of the antenna, distorting the radiation pattern and providing an easier path for ground noise pickup.

2.) I'm going to mount the dipole perpendicular to the PCB to keep the orientation of the fields the dipole can pick up perpendicular to any noise the esp might generate. To do so I have a cutout at the end of the PCB into which I'll insert the dipole. Then I'll solder onto the SMA pads on both sides. So I appreciate their presence in your design.

Joe

[captcha]

Thanks for the feedback, what I've got so far is:

> issue an updated version of the existing dipoles design that
    - has a bend radius instead of a 90 degree angle
    - reduces excess area on the SMA solder pads
    - has a 50 Ohm impedance for the short microstrip feedline between the SMA connector and the dipole
    - is re-checked for resonance on 433/868/915MHz

> issue two new designs:
    #1: direct solder-on pads (through-holes)

    #2: with pads for u.fl connector


For all designs:
  - minimise overall dimensions
  - issue as:
         - three-combo (433 with cut marks for 868 and 915MHz)
         - two-combo (868 with cut marks for 915MHz)
         - dedicated 915MHz


Then some points of interest.

The vias as shown in ChemE's post with the picture of the surface mount board is intriguing but should also raise a concern with feedline purists. Vias and through-holes in feedlines may be helpful for rigidity, but may also cause unwanted detuning and reflections because they are 90 degrees to the signal path. From what I've read I'm not so much concerned with reflections but I know that something like a tuned stub can be used a as an impedance match. The thickness of the pcb even at 915MHz probably still isn't large enough to be a significant portion of a wavelength but it is 'sticking out' somewhat. For sub GHz designs I think we can put this one to rest and just use vias and through-holes in most locations.

I've noticed that the SMA footprint is somewhat on the large side. I would like to shorten it and thin it out a bit, but keeping in mind to aim for 50 Ohms track impedance. Joe, you mentioned that you use the SMA pads to solder your board at a 90 degree angle to it. Will it jeopardize this feature for you? Can you comment on the freedom I have to change the geometry of the SMA pads? I think it's probably best for me to work out the dimensions for 50 Ohms first and then present here so you can see what it's going to look like first.

For the SMA connector design, I'd like to round the beginning of the dipole element tracks from the feedline somewhat. Not so much for concerns of reflections but more for trace flexibility because of the relatively small and flexible pcb. I'll try the 'Trace Curvature Radius >= 3 × Trace Width' approach and see how that goes. This is obviously of no concern for the solder-on design but maybe for the u.fl connector. With regards to the latter, Mark, can you advise whether you want the u.fl right in the middle or slightly on the side with a bit of microstrip like the current design? I guess if it's in the middle the width could come down dramatically but you may want some holes next to the coax to secure it in place.

Meandering dipoles.. verrry interesting, but way down the list for me, haha.. Thanks for the link though.

Any other suggestions?

[captcha]

..as soon as I mount the antenna close to the espgw board I get significant noise.

Ooh, that's not good. Only on receive or on tx as well?

1.) I decided to use a balun with the dipole to try and get better noise immunity. I settled on the HHM1522A7 (https://product.tdk.com/info/en/documents/data_sheet/rf_balun_hhm1522a7_en.pdf) for 868 mhz.  The reason I think it might help is that without it the outer surface of the coax and the ground plane of the board acts as third leg of the antenna, distorting the radiation pattern and providing an easier path for ground noise pickup.

You got it. The GND tracks of your espgw board WILL form part of the antenna, the trick is to a) minimise the amount of noise the antenna picks up from your board when it's receiving and b) minimise the effect of a radiating antenna on your board.

Also consider 'unintentional' antennas such as powersupply leads and other wires. At these frequencies almost anything acts as an antenna. It would be interesting to see if you can get even better immunity by adding small toroids to wires and by metal-shielding your espgw board.

The balun you found adds some losses but definitely helps to isolate common mode current caused by the (hopefully small) asymmetry of the dipole. I think you're on the right path, I just chose not to include this component in my design. However, I'd love to see some numbers when you've finished testing with the balun. It could be a great addition for those who want to experiment with antenna designs and I think it's a worthwhile path of investigation.

2.) I'm going to mount the dipole perpendicular to the PCB to keep the orientation of the fields the dipole can pick up perpendicular to any noise the esp might generate. To do so I have a cutout at the end of the PCB into which I'll insert the dipole. Then I'll solder onto the SMA pads on both sides. So I appreciate their presence in your design.

Are you saying that initially the coax feedline was not straightened out but maybe somewhat curled up? Then what you're seeing is expected behavior. Dipoles are very prone to picking up noise within a short distance of it. Not sure if this is purely within the near-field or just within the first half-wavelength.

Another point of interest is that, even with an isolation transformer, if your coax is not straight and perpendicular to the dipole for at least a half wavelength, it will very easily pick up rf on the outside of the coax when the antenna is transmitting.

However, if you're going to mount your espgw right at the feedoint of the dipole you may be asking for trouble. Without any shielding, you're putting your board right in the 'hottest' field of the antenna. I'm not saying it can't be done but if you're already having noise problems *with* a feedline, putting the gw board right on top of the dipole may make it worse. For your sake I hope you can prove me wrong! haha.. :-)

[joelucid]

Ooh, that's not good. Only on receive or on tx as well?

I haven't looked at tx. It's the receiver sensitivity that I'm currently optimizing.

Are you saying that initially the coax feedline was not straightened out but maybe somewhat curled up? Then what you're seeing is expected behavior. Dipoles are very prone to picking up noise within a short distance of it. Not sure if this is purely within the near-field or just within the first half-wavelength.

I'm working on an enclosure for the GW (now that Tom's initiated me to 3D printing ). For ease of manufacturing at first I tried a dipole where the lower half is the circuit board (gnd plane as lower leg). That is very noisy.

A dipole connected with a very short feed line has the same problem to a lesser extent -and without balun that's kind of clear: even with the pcb perpendicular you now have an unintended L dipole which efficiently picks up noise.

Another point of interest is that, even with an isolation transformer, if your coax is not straight and perpendicular to the dipole for at least a half wavelength, it will very easily pick up rf on the outside of the coax when the antenna is transmitting.

However, if you're going to mount your espgw right at the feedoint of the dipole you may be asking for trouble. Without any shielding, you're putting your board right in the 'hottest' field of the antenna. I'm not saying it can't be done but if you're already having noise problems *with* a feedline, putting the gw board right on top of the dipole may make it worse. For your sake I hope you can prove me wrong! haha.. :-)

I've got the pcb partitioned so that the esp8266 is as far away from the antenna as possible and I've added just about every trick I could find in app notes on EMC - short of shielding (beyond via shielding) and doing a 4 layer board. We'll see how it turns out.

The power line is a problem since the enclosure is a cylinder with the pcb horizontally in the middle, vertical dipole. So the power cable needs to go down to the bottom parallel to the dipole. I've been told I can avoid any impact on the antenna by shielding the cable with ferrites.

Joe

[perky]

The single most important thing for reducing radiated EMI is ensuring that at the point of PCB exit there is a low an RF impedance as possible between the digital ground and chassis ground, and that every I/O signal that exits a PCB has an EMI filter (usually using a ferrite and a small capacitor, and often an ESD protection diode has enough capacitance so both EMC and ESD are covered). If a metal box is used as chassis ground this means very fat and short connections as close as possible to each I/O connector, even a few mm of thin wire will have significant impedance at GHz frequencies and this will allow noise to exist between chassis and digital grounds. The low RF impedance of ground and chassis eliminates common mode noise, the filters eliminate differential noise. The next thing is to ensure all I/O is, if possible, sited along one side of the PCB only to stop a dipole effect of the signals exiting the board (this includes the antenna). Of course you also need to properly filter power supplies and use pi filtering for any high speed switching parts like an MCU and the radio on board, and use low leakage and low stray capacitance layout techniques for switching regulators. It goes without saying a solid ground plane is a must to reduce impedance.

If battery operated then chassis ground == digital ground, in which case the position of the connectors all down one edge of the PCB will have a signifcant EMI reduction impact.

Mark.

[joelucid]

My enclosure is 3d printed, so there's no metal chassis.

From what I have debugged so far I think the culprit is the esp-12e module. It's built on a 2 layer board and the flash chip isn't separately decoupled. Decoupling caps don't go down down into the pF. The signal return paths for the flash seem pretty long to me since there isn't a ground plane. There's a shield on top, but that doesn't fully enclose the module due to lack of ground plane and the i/o pins aren't filtered.

On top of what I said above my new design filters all IO and power at both the RFM and the ESP, uses a almost completely continuous ground plane with just a couple of very short bridges and ground areas on both sides are heavily stitched with vias. We'll see if it's enough.

If it doesn't one could look at shielding the esp-12e or doing away with the module and integrating the esp8266ex directly on a 4 layer board with the rfm. Any other ideas?

Joe

[perky]

That theory fits, it appears to be near field (primarily H-field I would say) coupling with your antenna which is what you'd expect if there are large current loops going on.

Looking at the layout of the esp-12e it does have a guard rail all the way around the can (tacked at every pin location) tied to ground at one point. That's not ideal but not a disaster, it'll still work as a shield especially to E-fields. However as you say the underside is essentially full of holes.

Do you fit other components on the opposite side of the esp-12e? Can you put a solid ground plane underneath it somehow?

Mark. 

[joelucid]

Do you fit other components on the opposite side of the esp-12e? Can you put a solid ground plane underneath it somehow?

No - the ground plane on the bottom is almost solid. The top side also has lots of copper and its all via stitched together. I'll attach the current design. Disregard the sizing of the pass thru caps, I haven't done those yet.

Joe

[perky]

Well I would say you've done just about all you can for shielding of the esp-12e . My guess now is that the proximity of the antennae means the rf transmitted from the esp-12e is de-sensitizing the RFM69HCW. Since it's OK with the RFM69HCW antenna some distance away it doesn't appear to be a lack of shielding over that. So can you turn the power of the esp-12e right down so all the rest of the logic being exercised but just not transmitting much rf, and does that help?

Edit: Maybe it's worth putting on a 915MHz bandpass filter inline close to the RFM69HCW? Something like this:
http://www.digikey.co.uk/product-detail/en/johanson-technology-inc/0915LP15B026E/712-1122-1-ND/1560951

Mark.

[joelucid]

So can you turn the power of the esp-12e right down so all the rest of the logic being exercised but just not transmitting much rf, and does that help?

Interestingly noise doesn't correlate with Wifi activity of the esp8266. But it does correlate with activity on the flash SPI bus. If I just measure noise and tickle the ESP's wdt to avoid reboot there's little noise. But as soon as I call yield the ESP starts paging in program code from the flash and the noise is back.

This makes sense as the flash isn't properly decoupled and the flash signal loops aren't very low impedance. Any E fields will likely be shorted away in the new design. But the H fields continue to be a concern. I have some ferrite plates here so when its assembled I'll try shielding the H fields that way to see if it helps.

Joe

[perky]

OK. Since you don't use the SPI pins (you've just got AC terminators on them), maybe changing it for a ESP-12S and add the central ground pad is a possible way forward (this apparently has an "improved PCB layout" and "better radio performance"). Maybe this lack of ground plane was a real issue with the 12E.

BTW have you tried it without the termination components fitted? I'm just wondering whether this forces SPI currents outside of the module shield area.

Mark.

[joelucid]

maybe changing it for a ESP-12S and add the central ground pad is a possible way forward (this apparently has an "improved PCB layout" and "better radio performance"). Maybe this lack of ground plane was a real issue with the 12E.

Excellent suggestion!  I hadn't see this module. I'll get some to try out.

BTW have you tried it without the termination components fitted? I'm just wondering whether this forces SPI currents outside of the module shield area.

Yeah, that's the way it is now. The design I posted I haven't received yet so I don't know how it will perform. I did add termination components to the old gw on the SPI lines during debugging. First just 47pF to gnd without a resistor which made things worse (understandably since this is a RF short). With 200 Ohm added things were unchanged from without termination.

Joe

[ChemE]

...Velocity factor for a lot of coax is around 0.66 so that should be all you need to determine how long the feedline needs to be.

As an example for 433MHz:
- Full wave in free space: 300 / 433 = 0.692 m (69.2 cm)
- Half wave: 69.2 / 2 = 34.6 cm
- Compensated for 0.66 velocity factor of RG-174 coax: 34.6 x 0.66 = 22.9 cm

So doing the math for 915Mhz, we end up at 108mm for our feedline length.  It looks like these UF.L to SMA patch cables are cheap and readily available on Ebay but only in lengths that are multiples of 5cm.  10cm patch cables are $1.50 shipped from China for instance.  Is the patch cable being short or long by as much as 2.5cm anything to get excited about?  In the case of 915MHz the mismatch is only 0.8cm which I'm betting is insignificant but I would like to understand the theoretical implications since I'm attempting to make the lowest noise gateway that I can.

EDIT:
Looking at the pictures from the auction perhaps the error is even smaller than I thought.  The connector adds 10mm making this 2mm too long assuming that the connector counts.


EDIT 2: I suppose I could set my center frequency to 900MHz (11cm ideal feedline) and see if there is any statistically significant difference in RSSI when compared to a center frequency of 915MHz (10.8cm ideal feedline).

[captcha]

The connector counts.

Ideally you want the whole assembly mounted/screwed in/attached/stretched out and then measure the distance between antenna feed point and the radio's antenna port.

I've ordered some u.fl surface mount connectors and will position it right in between the dipole elements. That should minimise any additional paths from the connector to the antenna elements.

[joelucid]

Excellent suggestion!  I hadn't see this module. I'll get some to try out.

Got the components (including the ESP-12S) and the new boards and soldered everything together. These feed through caps certainly are a pain to solder. But the results speak for themselves:

Noise level (at 55khz settings) measured just tickling the WDT:

Noise level, Percent of detections
-107.5,  10%
-108.0,  20%
-108.5,  70%
-109.0, 100%

Noise level measured with flash SPI active:

Noise level, Percent of detections
-107.5,  10%
-108.0,  30%
-108.5,  90%
-109.0, 100%

So they are now really the same and both are great and similar to a dipole connected via long coax. Now the real work begins: identifying which of the many changes actually fixed the issue. And of course build the enclosure.

Joe

[joelucid]

For comparison here are the same measurements with dipole and l/2 feed line:

Noise level (at 55khz settings) measured just tickling the WDT:

Noise level, Percent of detections
-105.5,  10%
-106.5,  10%
-107.0,  30%
-107.5,  80%
-108.0, 100%

Noise level measured with flash SPI active:

Noise level, Percent of detections
-101.5,  10%
-105.0,  10%
-105.5,  40%
-106.0,  20%
-106.5,  10%
-107.0,  30%
-107.5,  90%
-108.0, 100%

As you can see the new GW is better even though the antenna sits directly connected to the board. Measurements with l/4 monopole over board yielded noise floors in the -90dBm area, so you can see how much of an improvement this new design yields.

VERY HAPPY! 

Joe

[perky]

Excellent results Joe! I wonder how much difference the ESP-12S made in the end.
Mark.

[WhiteHare]

As you can see the new GW is better even though the antenna sits directly connected to the board.

It would be great to see a photo of how you've connected the antenna directly to the board.

[joelucid]

It would be great to see a photo of how you've connected the antenna directly to the board.

WH, look a couple posts up - I did attach a photo. The circular PCB has a cut out into which the antenna fits. And I then solder the antenna on all four sides against the PCB.

BTW, captcha, as you can see the perfect antenna for my purposes would just fit into the cut-out and have one leg on each side of the antenna PCB so that one side doesn't have to pass through a via on the dipole.

Joe

[captcha]

Interesting idea..

I'm not sure if I fully understand, but wouldn't you always need a via on the parent board to get to at least one leg of the dipole?

I think we may be splitting hairs here with eliminating vias though, as they would be perpendicular to the dipole.

[joelucid]

I wonder how much difference the ESP-12S made in the end.

What I do know now is that the new board works equally well with either the ESP-12S or the ESP12E in the dipole configuration.

The ESP-12S works so well that even if I replace the dipole/balun with a simple l/4 monopole with board as counterpoise I still get fairly low noise and in particular no impact if I access the flash SPI. However with the ESP-12E there is some impairment with monopole when accessing the flash. However it's still no comparison to the effects on old boards - I might see noise at -102 dBm vs -88 dBm with earlier boards.

It appears that my new board is really good at isolating the antenna from any noise the ESP produces. But the ESP-12S does seem quieter. An interesting test would be to drop the 12S into one of the old noisy boards to see if that already fixes the issue - I don't want to solder these feed thru caps any more often than absolutely necessary ...

Joe

[joelucid]

An interesting test would be to drop the 12S into one of the old noisy boards to see if that already fixes the issue - I don't want to solder these feed thru caps any more often than absolutely necessary ...

And here's the answer. My old gw with l/4 monopole and esp-12e while accessing SPI flash:

Noise level, Percent of detections
-89.5,  10%
-90.0,  10%
-90.5,  20%
-91.0,  10%
-91.5,  40%
-92.5,  30%
-93.0,  10%
-93.5,  40%
-94.0,  10%
-94.5,  30%
-95.5,  60%
-96.0,  30%
-96.5,  40%
-97.0,  20%
-97.5,  40%
-98.0,  40%
-98.5,  40%
-99.0,  50%
-99.5,  50%
-100.0,  70%
-100.5,  30%
-101.0,  50%
-101.5,  80%
-102.0,  80%
-102.5,  40%
-103.0,  50%
-103.5,  80%
-104.0,  90%
-104.5,  40%
-105.0,  50%
-105.5,  40%

Same revision GW, same antenna, but now with ESP-12S, while accessing flash:

Noise level, Percent of detections
-100.5,  20%
-101.0,  30%
-101.5,  80%
-102.0, 100%

MUCH better.

Here's without accessing flash:

Noise level, Percent of detections
-100.0,  10%
-100.5,  20%
-101.0,  10%
-101.5,  80%
-102.0,  90%
-102.5, 100%

So all in all the 12S is a huge improvement. Flash access is no longer an issue. But my other improvements (a) enable using the 12E without impairment and (b) further bring down the remaining noise.

Joe

[joelucid]

I'm not sure if I fully understand, but wouldn't you always need a via on the parent board to get to at least one leg of the dipole?

Look at the attached. The balanced output from the balun just goes completely symmetrical to both sides of the dipole. But then one side needs to pass back through the via to the other side on the dipole.

The other thing I'm thinking is to widen the copper legs of the antenna to the entire 5mm or so width I have. That would widen bandwidth quite a bit. If its wide enough one model might work for both 915 and 868 mhz.

Joe

[WhiteHare]

Is there any downside if I were to solder a Moteino directly to your antenna PCB without using a feedline?

In an attempt to answer the question empirically, I soldered it directly to the RFM69HW module on an experimental node.  In case others here are curious: it does work, and it gets better RSSI than a regular monopole.  Hence, probably almost any mote could be adapted this way.

[Edit1: I had filed down the edge of the PCB dipole antenna a little so that the trace pads were right on the edge, and my original plan was just to solder bridge the dipole antenna directly to the corresponding pads on the RFM69HW module.  However, the RFM69HW module had a thinner PCB, and for other reasons also, that didn't quite work out.  So, on this first attempt, I resorted to using a bit of wire to make the connections.]

[Edit2: Having tried it the brute force way, the amount of improvement seems a bit underwhelming.  It leaves me more motivated now to try it the proper way using a 1/2 wavelength of coax (as discussed above).  ]

[ChemE]

Did you get anywhere with rev2 captcha?

[captcha]

I've received the u.fl surface mount connectors and a couple of pigtail leads. I've redesigned the pcb layout to make them a whole lot smaller but haven't yet send them to the fab as I haven't yet made sure the diptrace library footprint for the u.fl connector is indeed covering all pads. I'll print them out on a laser printer next week and see if that lines up.

[TomWS]

I've received the u.fl surface mount connectors and a couple of pigtail leads. I've redesigned the pcb layout to make them a whole lot smaller but haven't yet send them to the fab as I haven't yet made sure the diptrace library footprint for the u.fl connector is indeed covering all pads. I'll print them out on a laser printer next week and see if that lines up.

Sorry if I'm out of touch, but are you thinking of having the u.fl connector on the dipole PCB with the pigtail cable soldered to the Mote PCB? 

I like the idea of having your dipole connected via cable and u.fl is a convenient and proven connector, I'm just a bit confused by your comment.

Tom

[captcha]

Sorry for the confusion.

As per your post, the idea is indeed to have a surface mount u.fl connector on the board, right in between the two dipole legs. For testing I will need some pigtail leads cut to the appropriate lengths so I can hook them up to a vector network analyzer. Once that's done I'll post the PCB dipole designs back on OSH Park. Those pigtail leads will of course not be part of the design, nor will the u.fl connectors; it will be just the dipole with the smd footprint to hold the u.fl connector, so you'll have to source those connectors and connecting leads yourself.

This is the u.fl connector I'll use to test the new dipoles:

http://www.ebay.com.au/itm/351802783571


This is a preliminary design of the dipole:


There will also be a 433MHz flavour and I'll also do up a design that has two through-holes 0.1" apart (instead of a u.fl connector) for soldering on a cable directly.

Hope this clears things up.

[ChemE]

What dimensions are you at right now out of curiosity?

EDIT: Upon closer inspection there is still a good amount of wasted PCB at the ends.  Why not move the outer-most holes towards the center a little and shave off that unneeded PCB beyond the dipoles?

[captcha]

The outside board dimensions as shown above are now 14.7mm x 11.4mm (5.8" x 0.475"). But please bear in mind these are oversized and not tuned at all for any frequency. I have deliberately extended the elements by a few cm to make sure I've got more than enough when trying to cut them to resonance for the various frequencies.

I'll reiterate: do not use these dimensions and expect that the v0.2 prototype will work. This is definitely a work in progress.. 

As per your suggestions for moving the holes... yes, that can definitely be done. Let's see how small we can get them.

[ChemE]

From the looks of it, it won't be too difficult to slide the holes up closer to the traces and shrink the height to 0.25" or pretty darn close.  Perhaps shrink the diameter of the holes if the board is still taller than 0.25" even after things are positioned more tightly.  I'm also thinking of trying these out with https://www.pcbs.io/ .  Four 915MHz dipoles for $5 shipped sounds pretty amazing to me.

EDIT: In looking again at your PCB on OSH Park, it looks like you could get pretty close to 4.5"x0.25" with the 915MHz u.fl version which would be $4.50 shipped for four!

[joelucid]

Btw if you make the dipole wider it can be shorter and will be more wide-band. See http://www.antenna-theory.com/m/antennas/broaddipole.php. For my gw I've got a 9mm wide dipole in the works which might cover the whole 868-925 MHz range. Which would be nice.

[ChemE]

Very interesting page.  Obviously the traces could be a good deal wider and the board still come in at 0.25" tall.  If this resulted in a still shorter PCB with better bandwidth so much the better!

[captcha]

I've also been wondering whether to go for 2 ounce copper (resulting in a pcb trace of 70um [2.75 thou] tickness). The skin effect at 433MHz is just a mere 3.1um and only gets thinner as the frequency increases. It won't help broadbandedness so much but it does reduce ohmic resistance and may help make it a bit more rugged.

[WhiteHare]

Will these 433Mhz antennas work just as well with LoRa spread spectrum?  I notice that AI-Thinker 433Mhz LoRa module are now nearly as inexpensive  as Hope RF RFM69 modules. 

Example:  https://www.aliexpress.com/item/LoRa-SX1278-433M-10KM-Wireless-Spread-Spectrum-Transmission-Module-Ra-02-SX1278-IPEX-Socket-DIY-Kit/32808223139.html?spm=2114.01010208.3.25.DJ5ts6&ws_ab_test=searchweb0_0,searchweb201602_5_10152_10065_10151_10130_10068_5160013_436_10136_10157_10137_10060_10138_10155_10062_10156_10154_10056_10055_10054_10059_100032_100033_100031_10099_10103_10102_10096_10147_10052_10053_10050_10107_10142_10051_5150015_10084_10083_10119_10080_10082_10081_10178_10110_10111_10112_10113_10114_10181_10183_10182_10185_10033_10032_10078_10079_10077_10073_10123_10120,searchweb201603_4,ppcSwitch_4&btsid=ed4c8f9a-190b-42e9-af5a-f927d136260d&algo_expid=e182d4a4-976f-44b4-8e7f-c164ffe2d510-3&algo_pvid=e182d4a4-976f-44b4-8e7f-c164ffe2d510

[captcha]

Just to keep everyone in the loop, I have just sent off a Rev2 of the 433MHz SMA dipole and a 915/868 U.FL dipole to OSH Park for manufacturing.

433MHz SMA dipole changes:
- increased the dipole element diameter from 1mm to 1.23mm
- pcb traces from SMA connector to feedpoint of the dipole has been replaced with curved traces with a bend radius of 3x the trace thickness
- modified the footprint of the SMA connector pads so they are now 50 Ohms Z.
- changed the center feedpoint of the dipole to be the middle of the board (not the SMA connector)

915/868 U.FL dipole:
- U.FL connector pads are 50 ohms Z
- trace thickness 1.52mm

These versions are deliberately 'too long' so I can hook them up to a VNA and cut them to resonance. Once that's done I will adjust the design for the correct lengths and post the link to the final versions on OSH Park so anyone can order them.

[captcha]

Sorry for the late reply:

Will these 433Mhz antennas work just as well with LoRa spread spectrum?

I believe so. Spread spectrum centers around the middle frequency that these boards are designed for.

[ChemE]

Thanks for your efforts captcha.  I've been holding out ordering my dipole until this improved design is ready for action so I'm glad to hear that it is mere weeks away now.

[captcha]

Thanks for your patience.

I had quite a few distractions and even lost interest when I wasn't getting anywhere with some of my other projects. Anyway, we're back on deck!

Yeah, it would be nice to have a complete set of dipoles for the three ISM bands with some common connection options (direct-solder-on, SMA and U.FL).

For now I haven't designed the direct-solder-on boards as I'm still contemplating whether or not to just use the SMA board for direct-solder-on; it already has large-ish exposed pads and it shouldn't be too difficult to solder the 3mm dia RG-173 onto it. Initially, with the solder connections I was thinking of re-using the U.FL board and putting in two through-holes instead of the U.FL SMD pads. That would make the board the same size but I'd still have to do the whole cut-to-size thing again because there will be slight differences in length from the connector pins to the dipole elements.

I think if most are happy with the SMA design we can safely say that this will behave fine with a direct-soldered-on bit of coax.

[Felix]

captcha,
I think the u.FL makes no sense on the dipole, SMA is more than appropriate.
IMO the u.FL would be more appropriate where space is very limited, and only from a PCB to panel inside an enclosure.

[ChemE]

Rather than the usual OSH Park, if you also upload your new design to PCBs.io and share it you get a portion of our purchase as a product credit for yourself.  Seems fitting that we could give a little back to you in this way.

[ChemE]

Okay, I got impatient waiting for captcha's revised design so I decided to take his initial measurement of 6cm for trace lengths and design from that detail alone.



I shared my design on PCBs.io since they are so impressively cheap and ship free all over the world: https://PCBs.io/share/zd3Ya
And the edge-mount connector I designed with: http://www.mouser.com/ds/2/238/consma003.062-273742.pdf
They sell 4 copies making them $1.17/each after your area discount but my last two orders came with 5 PCBs each making these $0.94 each if they keep up that freebie.

I had to get a little sneaky with the SMDs to get a standard RP-SMA connector to mount on the board and still have room to squeeze the dipoles into a 0.25" board width and keep all traces 40 mils away from the board edge.  I have a screen grab from Eagle below showing the pins and where they should land.


Going up from 0.25" drives the cost up rapidly so I wanted to avoid that if possible because I really want these to be super dirt cheap.  I went ahead and ordered a set of four and I'll be able to tell if my RSSI from nodes goes up using this as opposed to a 1/4 wavelength monopole.  I don't have a vector analyzer to be able to optimize the dipole lengths but if anyone in the US does, I can send you a spare copy for testing.

A question for the RF blackbelts: Are there any dire consequences to the choices I made in my design to squeeze this down to 0.25" board width?  I'm by no means an RF noob much less guru.

[captcha]

Great to see you're having a crack at it. You'll find it's not rocket surgery and once you have some working prototypes it's a super feeling that you're getting these awesome results with something you designed yourself.

I am in the process of making a video where I have done a shootout of several omnidirectional antenna designs and the clear winner is the dipole. I think this is purely because it has balanced radiating elements.

Thanks for your earlier post for mentioning to share the designs on pcbs.io and get a financial kickback. My main reluctance is that it's a different PCB manufacturer and I cannot assume that they will be using the same pcb material and manufacturing process as OSH Park. As you may know, the dielectric constant of the FR4 greatly affects the length of the copper tracks making up the elements and I would have to do another run of prototyping and testing before I would be confident sharing it with the rest of you.

With regards to the pads holding the mounting pins for the SMA connector, make sure that the right-most of the three pads (seen as an unconnected island) is not too close to the radiating element. There is a ratio to maintain 50 ohms impedance on PCB tracks. This also applies to the spacing and pcb track with of the center conductor in relation to the adjacent GND pads. Other than that, it looks fine to me. You've got a revision number printed as well, which helps you identify which board is which. I wouldn't worry about right-angled tracks either. I think you're on the right track (pun intended) to make a very useful and small antenna. Good on ya.

One thing I would include on your prototype versions, is to put a scale (like a small ruler) near the end of each track. That way, when you trim the ends for highest RSSI (or lowest SWR if you have the equipment) it's much easier to make both elements the same length.

Looks like you're not going u.fl after all??

[ChemE]

Thanks for the feedback captcha and good video showing the sticks and range testing.  I'll be doing something similar in East Park once my boards arrive!  I am adding scale marks every 25 mils on rev2 which I'll be ordering this weekend (for $5 and less than 2 weeks there is little penalty to fooling around).  @PCBs.io is on these forums so we can easily find out what board material they are using.

I'll have to look into the needed spacing more to keep 50 ohms and rev2 was going to use elements the full board width to try to get better bandwidth and perhaps be shorter because of their wide width.  I guess I'll just Dremel off that pin rather than rely on the soldermask.  I'll try it both ways since all the pieces parts are so cheap.  I was wondering if there is any benefit to placing a larger pad beneath the feedpoint to connect those two ground pins and stitching it to the top pad as you did in your design.  This way there is a ground plane beneath the feedline up until it enter the element.

[ChemE]

Looks like you're not going u.fl after all??

I think Felix is right and on a PCB dipole we don't need such a small delicate connector.  I did get some to solder onto my Motes and down the line a custom little Mote-clone.  They certainly make sense there where we want to keep our projects small but not lose the ability to go with a superior antenna if needed.  I initially wanted to get away from the SMA because I thought it was going to cause the board to be wider than 0.25", but if I can keep the robust connector on a tiny dipole then perfect.  I'm planning on also getting your design from OSH Park since they are the benchmark.  My own results won't be terribly useful to the wider community unless they are compared to the standard.

[ChemE]

With regards to the pads holding the mounting pins for the SMA connector, make sure that the right-most of the three pads (seen as an unconnected island) is not too close to the radiating element. There is a ratio to maintain 50 ohms impedance on PCB tracks. This also applies to the spacing and pcb track with of the center conductor in relation to the adjacent GND pads.

Okay from what I can Google up the spacing for a coplanar waveguide on FR4 is a 40 mil wide conductor with ground planes 19 mils away from each side and a ground plane beneath. Ideally the side ground planes are stiched to the bottom ground plane with lots of vias.

If I am extending these concepts correctly, I want to make sure the end of the rightmost SMD is exactly 19 mils away from the polygon coming off the feedline (center pin).  And I should connect the bottom SMDs together and stitch them to the top outer SMDs so that this whole area becomes a very short 50ohm coplanar waveguide.

Is that what you meant?

[PCBs.io]

Thanks for the feedback captcha and good video showing the sticks and range testing.  I'll be doing something similar in East Park once my boards arrive!  I am adding scale marks every 25 mils on rev2 which I'll be ordering this weekend (for $5 and less than 2 weeks there is little penalty to fooling around).  @PCBs.io is on these forums so we can easily find out what board material they are using.

I'll have to look into the needed spacing more to keep 50 ohms and rev2 was going to use elements the full board width to try to get better bandwidth and perhaps be shorter because of their wide width.  I guess I'll just Dremel off that pin rather than rely on the soldermask.  I'll try it both ways since all the pieces parts are so cheap.  I was wondering if there is any benefit to placing a larger pad beneath the feedpoint to connect those two ground pins and stitching it to the top pad as you did in your design.  This way there is a ground plane beneath the feedline up until it enter the element.

Sorry I missed this post. Our PCB material is FR4 with Er value normally between 4.3 to 4.9.

If there is sufficient demand/interest, we are not opposed making available a PCB antenna with edge connector kit.

Thank you,

Leo Wolfe
PCBs.io

[PCBs.io]

Okay, I got impatient waiting for captcha's revised design so I decided to take his initial measurement of 6cm for trace lengths and design from that detail alone.



I shared my design on PCBs.io since they are so impressively cheap and ship free all over the world: https://PCBs.io/share/zd3Ya
And the edge-mount connector I designed with: http://www.mouser.com/ds/2/238/consma003.062-273742.pdf
They sell 4 copies making them $1.17/each after your area discount but my last two orders came with 5 PCBs each making these $0.94 each if they keep up that freebie.

Five boards shipped Tuesday! 

[ChemE]

Thanks Leo!  And I've got 8 or 10 boards on the next panel too (with smarter silkscreening for tuning the traces).  Gonna have us an antenna shootout once those ship!

[ChemE]

ADMIN UPDATE: The following posts and dipole work by ChemE are kept for reference. But his latest dipole updates and design files are at this new thread.

First things first; major credit to captcha for putting out enough excellent information that I could get going on modifying his design!  I very much appreciate the foundation you created since my RF skills are non-existent.

Ok, I still have some tuning and experimenting left to do on this project and Luka has asked for a 433MHz version, but I'm going to call my smaller cheaper PCBs.io antennas an absolute unqualified success.  My testing method is to place a node sending a packet every 8s at 300kbps about 50' away broadcasting at -18dbm (power level 0) with a few walls in the way and then listened for 20 packets.  I sum the RSSI of those 20 packets and base my results on the average of those 20 RSSIs.  Captcha has already done quite extensive work to show how his PCB antenna compares to 1/4 wave monopoles, eBay purpose built antennas, etc, so I'm being lazy and just comparing mine to his. 

Captcha's average RSSI =         -91.65dbm
ChemE rev2 trimmed (too far):  -88.90dbm
ChemE rev2 untrimmed:           -84.5dbm

I don't have a VNA nor do I plan to acquire one so based on what I can test, these little babies are more sensitive than the baseline design (no idea why that would be the case) and importantly they are dramatically cheaper which was my goal.



You can order my 915MHz version here ($4.79 for 4 copies shipped anywhere in the world): https://PCBs.io/share/4XRDg

EDIT: Added the eagle files which confusingly are named rev3 but they are the correct ones which correspond to the board I tested and you can order from PCBs.io.

[Felix]

Awesome, just ordered some 

[ChemE]

Cool, glad to hear that Felix!  A note to anyone trimming these, the two radiating elements are the full width of the board (less the edge offset, so 220 mils) so a small scribe mark with a razor blade might not be sufficient.  I had best results getting a different RSSI by trimming the antennas with some heavy duty diagonal cutters that could clip all the way through the PCB and just physically shorten it.  At first I scored all the way down to the 20th silkscreen mark and wasn't seeing any real changes in RSSI before I busted out the dikes and chopped them and the RSSI dropped like a stone.  Luckily I have 5 copies of the board to make stupid mistakes on!

[captcha]

Excellent work Mr ChemE ! 

I think you did really well and have the results to show for it. I can very much relate the to the feeling of accomplishment when you've designed something from scratch and end up with a solution that is working the way you intended it.

Thanks for the kind words, but remember, I'm also standing on the shoulders of giants, haha..

As you mentioned, these PCB dipoles are just so much better than most cheapo vertical antennas you can get on eBay because there is a purposely designed symmetry between the two elements that make these antennas work very well. Which reminds me.. I still gotta post my antenna shootout video on youtube..

[Kylix]

Any chance to have a 433Mhz version of the antenna? 

[ChemE]

Awesome thanks Felix! I spaced the trim marks 50 mils apart from one another and the first mark is 25 mils inside of the edge of the resonator (true on both sides).  I used captcha's lengths of 6cm for the 915MHz antenna and 13.3cm for the 433MHz antenna.  From my R6 I soldered a u.fl connector and then used a 10cm pigtail to be between the u.fl connector and the RP-SMA edge-mount connector soldered onto the PCB antenna.  To keep things as tiny as possible, one of the ground prongs of the RP-SMA connector is isolated from the ANT resonator by just the solder mask.  There were some questions as to whether or not that was too close to cause problems but my results would indicate that this concern is more academic than real.  This all means the edge mount connector isn't quite in the center of the board; it is shifted right by 50 mils.  The real mirror plane of symmetry is centered between the two resonators.  Hopefully that give you enough to go on but if there are any any vagaries, let me know.

EDIT: Attached are my 433MHz Eagle files.  I got the board lengthened, the resonators stretched and the silkscreen shifted correctly.  The only thing left to do is place the edge mount connector in the right spot and then tune the gap between the two resonators; they look too close together.

[SabineT]

A version for 868MHz would be fine as well.
I'm in Europe, so I have to use 868MHz ;-)

[ChemE]

SabineT: No problem, I'll design one this afternoon.  It is entirely possible that the 915MHz antenna will work excellently with a 868MHz radio.  Using wide resonators is purported to make the antenna wideband; perhaps enough so that one design will cover both bands.  It is a cheap experiment to test the 915MHz antenna vs. the soon-to-be-designed 868MHz antenna if you have the right radio.  All my radios are 915MHz.

I'm in Europe, so I have to use 868MHz ;-)

Transmit frequency yes.  But the 915MHz antenna might be wideband enough to still get great reception at 868MHz even though its length was tuned to 915MHz.  Regardless, making a 868MHz design was trivial: https://PCBs.io/share/46L3w.  This will cost $5.11 for four copies shipped anywhere in the world.  Please let us know how these work.  I'm going to order a set and test this antenna with my 915MHz radios to see how it compares to the 915MHz antennas.

[SabineT]

Regardless, making a 868MHz design was trivial: https://PCBs.io/share/46L3w.  This will cost $5.11 for four copies shipped anywhere in the world.  Please let us know how these work.  I'm going to order a set and test this antenna with my 915MHz radios to see how it compares to the 915MHz antennas.

I ordered 4 copies right now. Thank you!

Regards, Sabine

[Felix]

ADMIN UPDATE: The dipole research by ChemE and related posts above are kept for reference. But his latest dipole updates and design files are at this new thread.
Thanks ChemE for all the effort, and of course to captcha for his research and valuable input!