Second build sold out. Message will appear here when store is ready for third build ordering.

Exact cause of Ethernet carriers (13 MHz and up every ~60 kHz)

This discussion was created from comments split from: Connecting Kiwi via WiFi.

Comments

  • Has anyone worked out if having the Ethernet live at a hardware level (I.E. lights on) is what causes the spikes or is it adding a specific type of wiring that allows the low level spikes to be picked up? if it is wiring could there be a solution like "Wire grade X, only pins 1,2,3,6 connected, conductors less than 150mm"
    I have so many spikes from other devices here (some of my own) I'm reluctant to spend too much time on USB wifi solutions if the benefit would be invisible.
    Could there be a way (assuming the pins are unused) to capacitively couple the spare Etherent lines to ground to reduce pickup or radiation, or to have a patch cable where 4,5,7 and 8 are cut at one end and terminated before the Kiwi?
    This is off-topic, maybe I'll do some tests, if I make any useful observations, start an optimising Ethernet thread.
  • jksjks
    edited August 2019
    There was a long thread about this back in 2017: http://forum.kiwisdr.com/discussion/comment/2531#Comment_2531
    Bonnie in particular indicated the problem was inside the Ethernet switch. And G0LUJ shows how equipment changes and relocation can help quite a bit: https://www.g0luj.com/?p=22

    I run my small switch from an external linear 5V supply, but still have terrible spurs. I opened the switch up and found a hefty 5 -> 3.3V switcher on the PCB. Someday I need to try replacing that with a linear regulator and heatsink. I'd like to know if this is really just SMPS related or something inherent in Ethernet switching. We have some evidence that it might be given that running the link at 10 mbps (see setting on admin page network tab) instead of 100 reduces the spurs (but does not completely eliminate them).
  • edited August 2019
    I did read that older thread, armed with that information, also due to other incomming noise, went short shielded CAT runs and fibre for any long bits.
    Issues here are mainly external to the Beaglebone/Kiwi, I can see at least nine off-set spurs at each step, a result of small houses close together. Peope, who fret about a single sub -130db spur are lucky.
    A few here are from my house (CCTV cameras and NVR), most are not, they are actually present at the antenna coming from neighbouring houses so make fine tuning a tedious task.
    I swapped from one LZ1AQ loop which has about 15m of overground shielded CAT5 feed over to the one with maybe 12m CAT5 mostly through 10m/10mm ? buried copper microbore.
    Some were better, maybe one was worse. I know throwing CAT5 antenna FEED in here may confuse but it's actully quite good at rejecting external noise pickup so helps spot internal spurs



    I must fire up the two Kiwi's in a box with a router/wireless bridge, see if they are worse, I did check quickly and didn't spot anything obvious, the leads in there are maybe 10cm and one aim was to address the chance of spurs by have a short ethernet and then break out to WiFI (when I can get the bridge to actually bridge).

    As dropping the clock frequency to 10mb made a noticeable difference at some locations I wondered if anyone considered breaking the ethernet specs to try and slug the signal even more, I.E. add enough filtering that the 100m Ethernet limit is dropped to say 20m@10mb but the radiated spurs are reduced to a similar degree. If that works we could end up with a "ethernet taming" PCB on tindie for SDR use.
  • With the Router (RB931) in the same metal enclosure as the Kiwi, wired, set to 10mb, there only poorly defined faint spurs.
    If the LAN ports are set to auto negotiation (100mb) there is an additional spur beginning to be an issue, if the WAN side is also set to “auto” another far more obvious/intrusive spur appears.

    I then added a USB-Ethernet adapter to the Beagelbone, as soon as the link lights came up new 60kHz spikes were very apparent, stronger than the router ones. That was also using a USB extension lead, the strength of those spurs surprised me.
    The new USB link did not show on ifconfig and was not passing data, the mere fact there is an active physical layer link, and the clock is running causes spurs.
    Each switch seems to run a distinct spur, some may overlap depending on the crystal accuracy.

    I tried a partially-populated Ethernet lead (browns and blues not connected one end) that made less difference than I expected (swapping it’s ends round made little difference) must revisit that.

    I then had another go at using the Mikrotik routers in wireless bridge mode.
    As long as the Kiwi (and Routers) IP addresses are set manually that works fine, I don’t remember having to go fully manual on the addresses before but may have.
    With no copper connection going out on the WAN side and very short cables on the LAN the Ethernet spurs were undetectable with a 50 Ohm termination or open SMA.
    Once the external antenna was connected I see spurs but external to the SDR.

    Later I turned off a couple of devices and narrowed down the spur origin. I can identify about 80% and they all disappear without an antenna connected (yes even if I crank up the waterfall).

    The images show blurred clock harmonics while passing data and steady at idle.



    What have I learnt?
    The spikes are not inevitable with Ethernet IF the Ethernet cables are kept short.
    Most modern houses will have some anyway.

    It may be possible to use off the shelf Ethernet transformers to break the long network cable runs that favour HF CM issues, but for me, right now I'll stick with a wireless bridge as one $20 router will do two Kiwi's. As for the spikes from other kit that needs more work, now I have a better fingerprint for the spurs I should have a shot at identifying the source and moderating those.
    Stu
  • Hi Stu,

    With the RF input to the KiWi terminated, what happens when you connect just the outer screen of your antenna to the outer of the KiWi SMA ?

    What happens when you connect the screen of your antenna cable to the metal screen on the KiWi Ethernet or USB socket ?

    Regards,

    Martin - G8JNJ
  • edited August 2019
    Hi Martin,
    Will have a look tomorrow.
    I've temporarly put the two Kiwi in a 2U aluminium rack case (from work scrap again) so everything is enclosed and grounded (via the SMA) and the top has to come off which is a bit of pain to change things.

    One thing I do differently is only ground the Kiwi through the SMA connectors, the Ethernet socket (or BBG PCB) is not directly grounded, I tried grounding it through the Ethernet socket while doing spur check yesterday and it seemed to give me some weird high noise condition but that turned out to be the station earth which I had to tighten (oxidation maybe). Once tightened it still seemed to offer at best, no improvement so Ieft it.
    I'm trying to avoid copper to the Beagle direct so the lack of direct static discharge route should not be an issue. The BealgeBone is also much closer to the base panel than the kit case (micro USB socket edge is about 0.5mm from the base - plastic shim between) I don't know if that helps noise at all, it was mainly done from necessity to get the lid on.

    So what you are asking
    1. Without the SMA grounded/earthed and the antenna terminated (50 Ohm?) take an (above ground) antenna shield to the SMA
    2. Same but shield to Ethernet or USB socket

    If that is right I'll try tomorrow.

    Regards
    Stu
  • I am using the Keebox USB wifi dongle (in stock on Amazon US) to completely eliminate the spurs associated with a wired ethernet connection to the Kiwi.
    Using it required the kernel upgrade to 4.4.155 described elsewhere in the forum and installation of the packages 'firmware-misc-nonfree', 'firmware-realtek', and 'connman'.
    The Kiw's external noise rejection is also helped by inserting an isolation transformer on the RF input, but discussing that deserves another thread.
  • Hi Stu,

    >
    >1. Without the SMA grounded/earthed and the antenna terminated (50 Ohm?) take an (above ground) antenna shield to the SMA
    >2. Same but shield to Ethernet or USB socket
    >

    Yes it's basically the same setup that you had for your test before, but with just the screen of the antenna connected to two different ground paths on the KiWi.

    I'm interested to see if the ground loop that is formed by the antenna screen connection influences what you are observing.

    Regards,

    Martin - G8JNJ
  • edited August 2019
    Hi Martin,

    I had some difficulty making meaningful comparisons as I had to use another Kiwi in the plastic skeleton case and I hit a bigger issue (that I had suspected may have a bearing).
    The distance above a good ground has greater affect on the spur height than anything I could do with more than one shield connection point.
    The act of pushing down at a couple of points was modulating the spurs as the Kiwi was sat on a folded paper guide above a well grounded aluminium case.
    I raised the kiwi up by 10mm (rested it on a plastic fan) and the spikes were about 10-15db greater. I had to break off for a call and my local noise source fired up so the general noise floor jumped, when I tried again like that it was more like 5db but still very obvious once the Kiwi was more than a about 5mm 1/4" from the base.

    I had demodulation set so that one spur was giving audio, as soon as there was any detectable RF I could hear it in the noise, before adjusting the waterfall for a visual.
    While moving around and trying to avoid grounds (I've spent the last year grounding everything) I worked out grounding is cryptonite to the spurs on a BeagleBone that does not have an external plane.
    I'd like to see someone with the proper metal case make up a grounded metal plate as close to the base of the BBG as possible without shorting anything and see if that makes a difference.

    This is comparing the same setup separated from the ground plane by paper or 10mm, not all the spurs are affected equally but two, I assume the local network induced spurs, are greatly increased.




    I'd like to say I am "confidence in my observations" but I am also cognisant of the findings of Messrs. Kruger and Dunning

    Regards
    Stu
  • Hi Rob,

    I want to try the USB dongle route but as many users may not want to dabble with Kernel upgrades I figured there was still mileage in reducing the normal Ethernet case.
    Here my $20 (for three wired devices) wireless bridge and decent earth helps greatly to the point I am pretty sure spurs see are all external.
    I'll go down this route until I have some measurements then follow your path.

    I do think having a working USB dongle is the neatest solution but also that there is something to be gained generally by using the spurs as indicators of installation efficiency.
    Sort the noise then go wireless, best of both worlds.

    Regards
    Stu
    (Martin has me all formal on the replies)
  • Notes to date.
    External spurs are common so no fault of the Beagelbone, internally there sould be perhaps a single clock harmonic (repeating every 60kHz or so) and the amount of pick up on the RX side depends on how well the tracks under the BB and incoming lead can radiate that spur back to the receiver and antenna. If the BB/Kiwi is in a conductive enclosure that has somewhere to dump energy and the only Ethernet lead is shielded and short (total for all connected leads from the switch/router) then the affect should limited. A quick way to see if the spur is local or remote is to zoom well in and pull the Ethernet lead for a second, the waterfall will halt but when it kicks back in you should see one spur (the local one) start narrow then go more blurred as the link is established and then data passes. I did screen dump that on another PC will add it later if I find it.

    The way to make the problem worst is to have the Kiwi/Beaglebone in a non shielded case, without an RF ground, connected to a long ungrounded antenna lead and miles of Ethernet the other side so that kit is inside a large antenna.
    If you do that and live almost anywhere in the developed world you'll see a local spike beside many other from unrelated networked kit.

    Looking at the schematic for the Ethernet RJ45 socket I see it has common mode mitigation on the TX/RX lines but the spares are just shorted then fed to a 1000pf cap via a 75 Ohm resistor. I figure that could be improved for this specific RF use by the clever RF guys out there....



    As a real quick test I put an air spaced variable cap across the unused pairs (using a POE injector) first hit I was able to reduce the local BBG spur going out of the BBG by about 5db (I assume at just one frequency), makes me think there must be a way to take that energy and dump it better than the stock BB socket does.
  • I suspect that part of the problem is that the values of filter components incorporated into the chassis mounted socket are not ideal for frequencies below 30MHz, which of course are the ones that are of interest to us. For example the 1nF capacitor = approx 5 Ohms at 30MHz. The transformers must also be very small in order to be built into the socket, so it's also likely that these don't work quite so well at lower frequencies.

    The configuration looks like a form of hybrid / balun who's main purpose is to maintain balance of the data pairs, minimise noise ingress at the frequencies used by the data stream (>30MHz) and help to comply with EMC regulations.

    I think that directly grounding the unused data pairs may help, but it could also introduce the possibility of unintentional ground loops.

    WiFi seems like the best option overall, as fewer cables (especially ones carrying high speed data) mean fewer points of entry for interfering signals.

    Regards,

    Martin - G8JNJ
  • edited August 2019
    Some interesting notes regarding the use of 75 Ohm resistors to provide "Bob Smith" termination in order to try and dissipate common mode signals, which I'd not previously encountered.

    https://electronics.stackexchange.com/questions/161436/why-is-bob-smith-termination-for-ethernet-recommended-if-its-wrong

    I couldn't figure out why 75 Ohms was used for this purpose as the characteristic impedance of CAT 5 is generally assumed to be 100 Ohms (although the characteristic impedance between pairs or bundles of pairs in a cable will be something different), but I would have have thought that any form or termination would probably need to be an exact multiple or sub multiple of this value.

    Seems this guy thinks so too.

    http://www.teli.us/EN_BSTermination04.pdf

    In other incarnations the 1nF capacitor is replaced with a much larger 100nF value. It seems to me that if all the 75 Ohm resistors are grounded at one point through a low value capacitor, which doesn't have a particularly low value of reactance, then all the lines will be 'floating' at some other RF potential above ground.

    This doesn't look like good practice and may simply been perpetuated after a flawed early implementation.

    Regards,

    Martin - G8JNJ
  • edited August 2019
    The size of the components occured to me, to be honest I didn't even think the socket was anything other than a connector. It was only after looking at Ethernet transformers and wondering what they had done after the nework chip here made me look. The centre tap transformers are generally used to recover DC from the pairs in some modes of POE (https://pinoutguide.com/Net/poe_pinout.shtml) I'm not sure what happens to the centres on the inboard side. From the brochure it seems common mode chokes are an option left off many connectors (that said some have them for all pairs).

    I agree Wifi seems like the more obvious choice but I don't like how we've gone from "no microwaves" to "full power through the whole house 24/7" in less than a generation, I avoid adding more items where I can.
    I'll probably dig a bit more just for the knowledge. Part of my motivation is to tame other network kit, some is OK, some absolutely terrible for SDR reception, after looking into this I might have a shot at working out why some clock harmonics are so bad.
    I'll probably end this thread here as the Kiwi is not at fault for being so sensitive and the BeagleBone is no worse than most kit, if I find a "Drop In" fix for external bad network kit I might post that as more people are probably causing the issue externally than seeing any internal pickup.

    Oh and I missed your second post sorry, interesting, a termination system I must use hundreds of times a day and knew nothing of it or it's failings.
    I did get the feeling there would be some way to better deal with the common mode, will have to try some options and see if there is a pin compatible socket with access to the termination values.
    In doing that I will probably destroy a BBG and realise how little I know about the physics, maybe next year...
    Another curious point is that C162 on the BeagleBone they talk about removing to deal with network at boot issues is labelled EMI on the schematics and is on the network clock, I won't be trying that.

    Regards
    Stu
  • Just some follow up on external devices.
    Once it got a feel for the idle/working spur signature I tried to identify any specific in-house spurs.
    As expected, the cheapest CCTV devices are producing the most visible spurs (except for whatever the closest neighbour is currently using).
    I have an SV3C bullet camera that is currently worst of the bunch, not only does it put strong spurs it puts out an extra "bonus" spur spaced by about 3750Hz up (every 60kHz repeating). I was aware this thing was bad but thought the FT-240 rings and careful arrangement of network leads and devices after that had tamed most of it, seems there was more to be gained. I swapped out the non shielded 3m lead +FT-240 for a CAT7 cable (ferrite on that made no difference) moved the large ferrite to the lead coming away from the next switch/hub, then another FT-240 on the far end. More ferrite on the (long) DC leads was noticeable too. That probably gained me about 10db.
    Peak hold in this grab shows the drop (the second high peak) the first missing peak is just from another cam that shares the switch - more ferrite on the DC lead shut that issue down.
    I hear thinking is terms of "sixth of a wavelength at the frequency of interest" is good guide, as soon as a lead gets to that it starts to be effective as an atnenna, I moved the ferrite to better break up the longer runs.



    I went round the spectrum as it's hard to get a feel for what is happening elsewhere while zoomed right in, 20MHz horrible and something else is peppering the band with spurs every 18.72kHz, I'm 99.9% sure that is next door as it looks similar to disconnected 'DSL but steady.
    One thing I have learnt, buy CCTV from brands that have a "reputation to lose", for example I have a D-Link that although worse image and value is solid as a rock an undetectable during these RF tests, if you do buy cheap add on about £100 for ferrite and a week for fitting ;-).

    BTW not expect replies just posting detail for anyone trying to reduce real local spurs at the antenna.
Sign In or Register to comment.