The forum is open to new posts.

Ethernet filters

Hi All,

I see that DX Engineering are now offering Ethernet filters.

"DX Engineering ISO-PLUS Ethernet RF Filters are very effective EMI Suppressors that are quickly and easily connected inline on Cat5e or Cat6 network cables. Their patent pending design provides common mode RF interference and EMI noise filtering of radio frequencies from well below 1 MHz to over 100 MHz. DX Engineering ISO-PLUS Ethernet RF Filters are bidirectional (input and output is interchangeable) and they have no effect on digital throughput; Ethernet data signal levels and speed remain unchanged."


Martin - G8JNJ


  • Nice catch Martin. The last word in the keyword list says "pair". And if you bring up the specifications tab it says "quantity: sold as a pair". That makes the price a little easier to consider.
  • I noticed the DXE devices when they first appeared - and have wondered about them. I did notice that they didn't support POE (hence the claim of galvanic isolation) which means that they would be of little help when it comes to quieting the oft-noisy IP cameras that are becoming increasingly common, nor would they help in preventing the conduction and radiation of power supply noise from a POE-powered wireless link radios (e.g. for Internet or "Mesh" operation) - or from the power inserters/Ethernet switches that power these devices, which are often bad or worse than the device being powered in terms of RFI.

    This makes me suspect that the DXE devices are simply back-to-back Ethernet transformers - with some care taken in the design to allow operation at Gig-E speeds.

    For quieting Ethernet connections I have, for some time, just used the "flat" Ethernet cable (often called "Cat-6" or "Cat-7", regardless of whether they really are...) wrapped around large-ish ferrite toroids. It would seem that one can wrap this "flat" cable rather tightly without upsetting its geometry as it will still work fine at Gig-E speeds when "spliced" inline with a Gig-E circuit, provided that one uses a Gig-E rated double-female RJ-45 connector.

    In testing, I've verified the following in a fixture that allowed the insertion of an Ethernet cable/device and inputted a signal on one side with a spectrum analyzer on the other side, terminating in 50 ohms:

    Using 10-12 turns on the following cores (approx. 33mm O.D., 23mm I.D., except where noted):

    - Mix 77: >=20dB across AM BCB and onto 160 meters (Fair Rite 5977003801)
    - Mix 31: >=20dB from 160 meters through 30 meters (Larger core - Fair Rite 2631803802 - I haven't found a convenient source for the 33mm O.D. in this mix)
    - Mix 43: >=20dB from 40 meters through 10 meters (Fair Rite 5943002701)

    For a particularly troublesome situation, three cores with 10-12 turns each - two Mix 43 and one Mix 31(or Mix 77, which works almost as well at 40 meters ) in series, but each separated by about 1.5 inches (3-4cm) to prevent cross-coupling seemed to work well from 160-10 meters with well over 30dB isolation (>50dB measured at some frequencies) and about 20dB to 6 meters. I've used a similar scheme with a Mix 77 core to keep my 630 meter and 2200 meter transmissions from getting into devices.

    Because these are simply common-mode chokes, they have no effect on (properly made) signals on the Ethernet cables themselves - and can be used to pass DC (as in POE).

    For higher frequencies (VHF, UHF) it gets increasingly difficult to effectively quash radiation on cable, but more common split ferrites designed for the frequency range, placed as close to the offending device(s) as possible, are the best bet.

  • As 98% of my Ethernet issues are from others they are a bit expensive to sprinkle round the neighbourhood, would be very interesting to try though.

    I have used "certified rubbish" CAT leads without the cover wound on ferrite torroids as last-foot leads, without the outer cover it is much easier to wind them on small forms and the lightweight conductors help assembly if only on a short patch .
    I used to avoid ethernet joiners wherever possible but after watching some (physical) networking guru on youtube show he couldn't do his impressive fan outs and flows without joiners I realised they were worth considering if decent quality.

    Will probably take Clint's build tips and stock some more serious chokes for when I move the Kiwi to a location worth the effort.

    I recently bought a Chinese PTZ IP camera, very quiet electrically (on HF anyway), had me double checking it was powered. Never thought I'd say those words, can only assume there was enough of a market to justify the right components, all other budget IP cameras I've touched are from poor to truly criminal.
  • I thought I'd try measuring them with the same fixturing I used to measure the CM rejection for currents,between the SMA ground and the PS/LAN side.

    Ordered but:
    " Part Number: DXE-ISO-PLUS-2 This part is temporarily out of stock, but you can order it now. Estimated ship date: 3/21/2019 "
    perhaps they are popular already?
  • For $US50 DX Engineering shipped me a pair of filters and two short cables to allow inserting them in-line. But before actually trying them out, I had to at least take one apart to see what they were. Essentially these are back-back HALO HFJ11-1G46ERL Ethernet filters, RJ45 connectors with built-in current and voltage baluns

    As you can see there is no DC connection between shields and there is the balance/symmetry of the transformer isolating shield and pair currents from appearing on the output lines. DXEngineering's page doesn't indicate what the test conditions were for their measurement showing > 20 dB CMRR across the HF bands. In use, the attenuation obtained will depend upon source and load impedances which aren't specified but were perhaps 50 ohms for their measurement. I'd expect typical long lines of CAT5 to show considerably higher impedance at some frequencies and so net reduction of CM current may not be nearly so great as their plot would indicate, at least not everywhere.

    They indeed will not allow POE since one of the two baluns is flux-coupled and is essentially a HPF with no DC response.

    From Martin's, G8JNJ's, schematic of the BB grounding it looks like these filters could potentially improve the noise floor for KiwiSDRs. If I'm reading it correctly, the CAT5 shield is connected to BB and enclosure grounds directly. Thus any CM current injected onto the CAT5 shield or lines and given a return path out the SMA isn't suppressed. As I've written before, I measured about 70 dB attenuation for this path - an isolated 0 dBm 50 ohm source placed across the SMA/LAN-end grounds resulted in about -70 dBm signal detected by the Kiwi. This implies to me that many of us may have our noise floors raised with spurious signals getting in by this mechanism. Since the ultimate floor of the Kiwi itself is about -157 dBm/1-Hz there's a lot sensitivity to see this kind of interference.

    But the LAN cable is not the only way for CM currents to be injected, the PS and even GPS cable are other possibilities as well. And I haven't even tried the filters out here and so it's too soon to judge, just thought I'd pass along these details since I was thinking about it.

    If you want to roll your own filter, Mouser sells them in singles for US$12.50 so DX Engineering is giving them to us mounted on a board in a little molded housing along with a cable for the price you'd have to pay for 4 connectors alone. However, if you're willing to make several yourself, at quantity of 25 they drop to US$8.57/per and getting the finished product from DXEngineering isn't such a good deal. I wonder if we'd see any improvement in a KiwiSDR if the RJ45 connector were replaced with one of these.
  • So DX Engineering is selling these in pairs, presumably to be installed on both ends of an Ethernet cable to prevent the cable from radiating and to help isolate the attached equipment from the common mode currents on the cable? In the case of the KiwiSDR, if only one of these is used at the KiwiSDR, it would work to isolate the KiwiSDR from common mode currents on the Ethernet cable, so only one of these units would be useful? If that were the case, it would be more effective than ferrite toroids?

    I guess I'm getting lost in this effort.
  • I could be wrong here but one of the more pressing issues with Ethernet is noise getting in, as well as out, common mode RF could be significant and may stop more things than are affected by radiated noise.

    I'm still a fan of taking up fibre as it's hard to induce anything in that even if you have 100m coiled at your feet. The only trouble I have with fibre is planning what items will have IP+galvanic connections and what just IP.
  • Hi Glen,

    Can you clarify a point. In a single DXE filter unit, are there two of the "circuit F" filters (that you showed in your post) connected back to back or does each unit only use one of them ?

    How are the ground connections handled and what are they connected to ?

    The capacitor values also look a bit on the small side for the LF bands ?


    Martin - G8JNJ
  • Yes, DXE sells them in pairs or packs of 10 (US$240). I suppose they are implying what you say, kill CM current in each direction. Protecting only the Kiwi and not the router/computer could be adequate for our use.

    Fiber is great but it doesn't solve the power supply or GPS cable ingress mechanisms. It's any-line-in/out to SMA/Feedline out/in currents we are trying to quench and we would like on the order of ~130 dB total isolation. Another 40-60 dB wouldn't be excessive but may be difficult to achieve. This is why I have been recommending a Guanella balun (current transformers) on all incoming lines so that the Kiwi is totally isolated from the world. It may take a couple of different toroid cores to cover the L-HF spectrum and each has to be big enough to get a few turns from every line in/out of the Kiwi, including getting connectors through.

    Finding a non-conductive fiber interface that doesn't need to have conductors added to support power for active circuits could be useful. I haven't looked and don't know what's out there. Anything we can do to reduce/eliminate CM current paths into the Kiwi may be helpful. The best solution I've found is a small active antenna, a dipole, situated right next to a Kiwi with an attached WiFi network dongle, the whole mess powered from a LiPo battery. This allows a feedline-less receiver with low self-capacity such that there can be very little CM current. With that it is possible to get to the next level of noise sleuthing to start to discover what near-field mechanisms may be contributing to increased noise/ingress. I'm pretty convinced that unwanted ingress into our systems is almost never actually by inverse-square law radiation. This means that once the antenna/receiver is truly only responsive to differential signals the remains are all near-field in origin. This is truly good news since it means they drop much more quickly with distance from the source. With a portable feedline-less Kiwi it is now possible to do a 'site survey' to not only discover the sources and possibly quench them at the source, but to find locations and polarizations which minimize them when it isn't possible to kill them where they start.

    This reminds me to restate one other discovery. I find that the two-wire alternate antenna input on the Kiwi is *much* worse than the SMA in regards to CM current impact. It may be 20 dB or more worse. I recommend never using it at all if one cares about sensitivity and noise floor. In fact, I've removed it from one of my Kiwis. I believe this is an indication that different ground current paths on the Kiwi PCB are involved when this connection is used and that they aggravate the situation. If you need to use an unbalanced non-coaxial fed antenna such as an end-fed wire, I suggest the absolute best broadband balun you can muster, flux coupled not a current balun, between the SMA and that antenna. Mini-CircuitsT1-6+ 1:1 transformer is one possibility that won't sacrifice Kiwi bandwidth. It perhaps isn't quite as good for the purpose as one that would work even better for MF-HF but it may help a lot. You are aiming at isolating grounds and minimizing inter-winding capacity across the transformer.
  • jksjks
    edited March 2019
    I believe this is an indication that different ground current paths on the Kiwi PCB are involved when this connection is used and that they aggravate the situation.
    Nope. Same as for the SMA. You have to look at all the 6 PCB layers with a Gerber viewer, because there are no PDFs of the individual layers: At that end of the board all 5 of the inner and bottom layer are ground plus the top layer copper pour. So plenty of continuous ground to go around. And of course all the ground vias at the connectors, edge stitching and everywhere else connect to all the layers.

    Thumbs up for using the MCL T1-6. I've used those here and they work great down to 10 kHz / Alpha reception range. Note also that the T-622 with its 3 trifilar windings also works to 10 kHz even though the published curves only show measurements to 100 or 250 kHz (can't remember). With 3 windings you can run two receivers off the same source with only the usual splitter loss. I'm using those currently.
  • Perhaps you misunderstood me. I am suggesting that simply moving an injection point a cm along the edge of the board or between the SMA ground and the two-wire ground results in a very significant difference in detection level. This isn't about whether grounds are continuous or not. Simply altering the current paths on a sheet of copper is changing what appears across the LTC6401. There is a different path and symmetry from different IR drop alone. "Ground isn't ground", it's got finite conductivity and at the levels we are dealing with current paths have to be considered because of IR. I invite you to try the experiment with a well isolated, low self-capacity current source.
  • At 10 GHz, sure. Not here. I'd like to see some actual measurements.
  • Just on the GPS ingress, use on of those car GPS repeaters.
    Very short (or highly filtered) GPS lead on the Kiwi with the extender TX in a box, very hard to unduce LF/MF through that (side observation from other test).
    I did find that here the GPS lead didn't induce much noise, surprised me as everything else here picks up RF for pastime.
  • John,
    You may be correct that this isn't the coupling mechanism, I haven't proven it to myself that it's not instead, say, a magnetic loop rather than straight IR drop. But it seems to me the important aspect is that I *did* measure a significant difference in the KiwiSDR response to injected RF ground current based on injection location. That this coupling exists, whatever the mechanism by which the preamp couples to it, may be of importance to any of us battling CM noise ingress. That a 0 dBm 50 ohm source of current produces a KiwiSDR response of around -70 dBm seems relevant to me anyway.

    And don't misunderstand me, I'm not at all convinced that the KiwiSDR isn't presently better in this regard than many or most commercial radios. I haven't carefully measured but my sense is that my 'other' SDR, an Apache Angelia board, is worse in this regard. We have a sensitive 'eyeball' and 24 bits down from -14 dBm OV is a long way down!

    Out of curiosity, I just performed a 4 terminal measurement of the plane resistance of a KiwiSDR PCB.

    I injected 1.5A at an SMA ground pin and C703 on the other end of the board. I measured 2.5 mV for an indicated 1.7 miliohms at DC. I didn't probe to determine how linear with position the resistance might be. If one expects that resistance will be no lower at LF-HF then 1 mA injected can be expected to produce at least 2 microvolts drop end-end. I don't know how that maps to the effective inputs of the preamps so I can't really say what that produces but if one gets 30% of that, the approximate portion the front end occupies, it might not be unreasonable to see > -110 dBm. OTOH maybe that coupling is 40 dB lower than this and below the -157 dBm/1-Hz floor of the KiwiSDR.

    This is all too back-of-envelope to mean much and it may be that there is a different mechanism but someone can pretty easily duplicate my RF current injected measurement to see if they observe the same thing I did. Maybe someone else will. I'll put further investigation of this into my too-long list of 'interesting things to do'.
  • Well maybe "someone can't easily duplicate my RF current Injected measurement" . I can't do it now !!
    Several weeks ago I measured similar numbers on two different occasions with one of my enclosed kiwis. I no longer know which unit it was. Trying again tonight with a bare kiwi not in an enclosure as well as with one that is enclosed I'm getting values that are at least 20 dB lower and it's unclear that the mechanism isn't just e-field coupling.
    Too many layers to unravel here so don't anybody else waste your time on it for now!
  • Does anyone know an effective broadband common mode filter for USB?
    I found that some USB devices produce noise and I was only mildly successful with ferrite chokes.
  • There are filters aimed at audio enthusiasts, can vaguely remember seeing some good reports on those when used with SDR (Jitterbug in search).
    from personal experience, mainly with the old softrock SDR's I found not all USB leads are the same, big "beefy" ones I assumed would have great shielding seemed to cause more noise in that use than some others that were thin and probably less material, it's also easier to put loops through ferrite if the lead is smaller. I tend to think in terms - move it away, reduce the length of the cable, more ferrite turns, try different cables (even the rubbish looking ones) ground devices in different ways see if that helps.
    Another long shot to consider is USB over wifi, try a USB-IP solution like VirtualHere but then make the network side of the link wireless, one USB device is free so it can be tested on say a raspberry PI zero wireless.
    I ran my old SDR-IQ on a PI Zero the other day, worked well. The license is tied to a physical host so I am reluctant to pay $50 to tie the license to a PI Zero but I might try it on a Linux machine later. I wanted to test some SDR's wirelessly over USB but even the FunCubeDongle is two devices.
Sign In or Register to comment.