The KiwiSDR 2 online store is open for orders! Please visit kiwisdr.nz
Please visit kiwisdr.com (documentation) and kiwisdr.nz (online store)
Halo Electronics ethernet filter - PCB design proposal
Hi Group,
I have 2 Halo Electronics ethernet isolators here. Rather than butcher them together with wires it would be nice to design a small PCB to for it. Also to design the PCB for a standard size diecast metal box.
Has anyone produced a PCB before for this job? Does anyone have mechanical suggestions for the board design?
I'm hoping to order another small PCB alongside these, the price will be low if anyone is interested?
My Kiwi has problems with ethernet "lines" this may help.
73s, Rob
M0RZF
I have 2 Halo Electronics ethernet isolators here. Rather than butcher them together with wires it would be nice to design a small PCB to for it. Also to design the PCB for a standard size diecast metal box.
Has anyone produced a PCB before for this job? Does anyone have mechanical suggestions for the board design?
I'm hoping to order another small PCB alongside these, the price will be low if anyone is interested?
My Kiwi has problems with ethernet "lines" this may help.
73s, Rob
M0RZF
Comments
Even though the CM rejection of a stock kiwi/BB LAN interface isn't bad, it's not enough to prevent typical noise currents (often a family of spectra with ~60 kHz spacing up in 20-30 MHz region) from degrading performance. Unless something else in the system is worse, these terms likely are hurting performance with simple installations.
Though addition of these filters may improve performance, I think it worthwhile to examine the situation to see how much isolation/CM_rejection is actually necessary with typical installations to remove the problem completely. The question revolves around another question "how much CM current is actually on a LAN|PS|Antenna line" and how does that map to coupling and the kiwi's noise floor?".
When I looked last, the isolating RJ45 was about US$12 in single quantity. A US$50 contribution to DX Engineering might avoid a butcher job and may help answer some of these questions, though more investigation is likely required.
But ham radio products attract large tarriffs so the DX Engineering parts will be expensive outside the States. Look at the price of Elecraft equipment in UK magazines and you will get a shock!
I tried all sorts of ferrite chokes. The problem is worst between 15-25MHz where nothing seems to make any difference. Because my KiwiSDR is powered from a solar panel, having a greedy WiFi link to avoid the cable is not attractive.
As I have the filters already, designing a board is not too onerous.
-Rob
You may very likely find that the problem is related to the antenna/SMA side of things. Unbalanced current injected by the imperfect LAN balance and finding a CM exit path out the antenna coax will very likely produce the results you are experiencing. The problem may better be solved by working on antenna transmission line balance. Try terminating the SMA at the kiwi and see if it changes the situation such that you see a coherent-interference free noise floor at around -122 dBm in a USB bandwidth.
If you can do it, try a low inter-winding capacitance 1:1 transformer at the antenna end of the feed line. The Mini-Circuits Tq-1 isn't too bad, though it still has about 1 pf coupling so is only good for an extra 30+ dB of CM rejection at 30 MHz. You'll know you have done the job well enough when you can short the antenna side and you see the kiwi noise floor the same as if you put a termination right on the SMA. Until you can do that, you risk having ingress from LAN and other sources degrading your receive system.
I just measured and compared a LTC6432 preamp, kin of the LTC6401 in the kiwi, using Mini-Circuits JT-1975+ transformers on both the input and output with a DX Engineering/Clifton Labs Gal-74+ preamp (4 devices) and ferrites that have a great deal more core mass. I did this on a good sized antenna in a suburban/urban environment. I have not seen any discernible evidence of problems related to the transformers. In fact, both preamplifiers look essentially identical except for the 8-9 dB difference in stage gain and lower noise of the LTC part.
Because this transformer has a better low end (40 kHz) I suspect it has even higher permeability ferrite than the T1-1 and is more like the T1-6. The latter have shown themselves to act a lot more like transmission line transformers and while they have better LF operation do not provide the common mode isolation of the T1-1 but for neither these or the JT-1975 has IMD been evident.
If you've seen a problem I'd like to know about it so that I can look even more closely. IMD values for the LTC preamp look pretty good and noise figure is significantly lower than the GAL-74+ preamplifier.
I tried 2 different buck converters which made no difference to the main spurii. The nature of the KiwiSDR with so many connections makes it hard to pin down where it comes from.
Making a PCB is not difficult as the CAD software has a defined footprint for ethernet sockets.
-Rob
This is particularly problematic when the RJ45 sockets incorporate an internal transformer / Balun network with "Bob Smith" termination, as they are not optimised for frequencies <30MMHz and the "Bob Smith" termination methodology is dubious anyway.
If I was designing an Ethernet filter I think I'd just use a bare bones socket and add external filter components that were properly optimised for the bands we are interested in.
Better still just use a USB WiFi dongle to provide the network connectivity.
The smaller Mini-Circuits transformers do produce IMD, as I found when I had to take them out of my home build IMD measurement rig.
However this only started to become evident at levels above about 0dBm, with no DC component present, as PoE adds further issues.
Regards,
Martin - G8JNJ
I added some of those here and felt that my MW and lower HF performamnce degraded. I split my current antenna 3-ways and a kiwi with and w/o to compare.
There is also this for review https://www.qsl.net/wa1ion/bev/bb_antenna_matching.pdf
A write-up on that may be found here: https://ka7oei.blogspot.com/2019/11/homebrew-construction-of-2-and-4-port.html
These should have no issue handling any amount of incidental RF energy that one is likely to see as similar devices are used in QRP transmitters - although it is important to make sure that very low frequency energy (particularly mains-frequency and immediate harmonics) be blocked from from any such ferrite device lest related modulation/IMD result.
This information is primarily for transformer-type splitters on unbalanced lines and as-described, they would not have particularly good longitudinal balance at the higher end of their range, requiring a bit of re-work. Perhaps symmetrical transformers and/or electrostatic shielding which can yield excellent results at lower frequencies (to, perhaps a MHz or so) but the latter can get increasingly difficult to do while minimizing roll-off at the top end of HF, particularly with the likely number of turns needed to allow LF (<30 kHz) operation.
73,
Clint
KA7OEI
Related to Martin's testing: perhaps the reason I haven't seen any issues yet is that my fixturing and testing has generally been at -10 dBm, 10 dB lower than his but about as high as any signals I see here or at a previous location 5 km from MW BCB. Certainly some users will see bigger fields so perhaps if I reconfigure my dual-tone system for another 10 dB, IMD due to transformers may pop out of the noise. I didn't do this previously because I was being very careful to be sure that the HP signal generators used as sources did not generate IMD in their output stage(s).
I've generally built tuned isolator/combiners at 10 or 21 MHz and introduced the two tones at .1 - 1 MHz spacing, enough to be sure phase noise in either the sources or the analyzer used to measure isn't an issue. I've also used a similar setup to measure with 100/100.1 MHz tones to check robustness against FM BCB IMD since these are some of the largest signals I see where I live.
I don't doubt that at some level small volume ferrite cores will saturate and generate IMD but with either Mini-Circuits or my own wound transformers and levels I have found to be typical of maximum signals I haven't yet seen a problem.
If I manage to reconfigure and remeasure I'll try to report the results.
My observations go along with what Glenn (N6GN) says in that the Ethernet signals are fast risetime with Volts of magnitude between the twisted pair wires. Common mode rejection is not going to be adequate enough to prevent a lot of stuff being seen by many antennas. Any really adequate filter will not let the data through. So, somehow, the answer is shielding which in many situations is practical to do.
73, Bob W7PUA
Thanks for posting. I find this kind of detail helpful. Your four measurements seem just about right. If I understand things correctly, your spectrum display has bins 30MHz/1024 wide so KTB should correspond to a display of about -130 dBm. Since you are showing -109 at 15 MHz (where the 'factory' -16 dB setting makes the Kiwi correct) indicating a NF on the order of 20 dB and a floor of about -154 dBm. This seems exactly right from my measurements. Do note that in operation with an antenna things can typically get 2 dB worse in the presence of big signals which exercise the top bit(s) of the ADC and the part's nonlinearites.
Because your antenna is small, I suspect that in upper HF the noise you are seeing with the preamp on is due to the preamp rather than to propagated noise, which is perhaps only 10 dB above KTB. It would be interesting to compare the top plot with that from a shorted loop, over the entire 30 MHz. I suspect there will be change at the low end but not at 30 MHz.
By my reckoning, a smallish, say 1m, broadband antenna whether loop or dipole, can't get us to propagated noise above 10-15 MHz with any known preamplifier architectures and components, at least while the sun is in low activity as it is now. It's for this reason I've been pursuing a hybrid antenna approach.
I also have a working 5722 vacuum tube noise generator that could serve as a calibration for the preamp at these frequencies. I will try to do that. -Bob
Yes, you've just described the essence of recent endeavors. It appears that a 1-2m antenna (I've been using a 2m dipole) and available components can get one to the propagated noise floor up to somewhere around 10-12 MHz. I think it doesn't change much whether one uses a loop (low impedance preamp) or a dipole (high impedance preamp) relative to device noise limitations. A broadband (unmatched) antenna in this range feeding one arm of a dipliexer combined with a larger matched antenna such as the modified biconical I described with another 15 dB or so gain (LTC6432-15 looks promising) feeding the 12-30 MHz arm appears to have a chance of getting the entire LF-30 MHz range covered sufficiently well to be limited by DX noise.
Use of balanced antenna structures, as opposed to monopoles over a ground/image plane, along with special attention to balun design (I've been using an active balun on the active antenna and a passive over the upper range) along with general care for ingress suppression of the type you describe for LAN/PS seem to make it possible to get the propagated-noise-limited SNR transferred to the Kiwi (or other receiver) over the entire VLF-HF range. All this is modulo keeping near-field noise sources sufficiently far from these antennas so as to not raise the floor. Because these drop off very rapidly with distance, 4th or 6th powers of D, I think this may be viable for even many residential environments.
Making all this work is a constant peel-the-onion operation of identifying and mitigating noise and IMD problems along the way but I don't (presently!) think it's impossible. I don't think it's possible with a single broadband antenna though. Generally speaking, above mid-HF a Kiwi needs some gain ahead of it and after a losslessly matched antenna in order to get to the desired noise floor.
A couple of us have measured several Kiwis with good agreement of results. See section 8 of Estimating_LF-HF_band_noise_while_acquiring_WSPR_spots I suspect that your 5722 measurement (watch out for turn on/off transients driving a DUT into non-linearity if you switch it rapidly) will agree. As noted before, NF measured at small signal levels will typically be a couple of dB optimistic in predicting kiwi noise floor in the presence of large signals.