Flatness of KiwiSDR response < 500 kHz?
I'm finally chasing down local LF noise sources. With virtually every grid load being a nonlinear device of some sort these days it seems almost hopeless, but I have to try.
One major offender in my shack is an APC BACK-UPS XS1500. Interestingly it's actually quieter when running on inverter and disconnected from the wall; the battery charger that cycles on and off to keep the battery topped off seems quite noisy. It switches at 71.5 kHz, and while I can see a line there in the KiwiSDR waterfall its harmonics appear MUCH stronger.
It could simply be antenna rolloff (I'm using an untuned G5RV for this testing). But I'd like to know the nominal frequency response of the KiwiSDR below 500 kHz or so; I don't have a calibrated signal generator to find out. Thanks!
From an hp 8657D coupled with a "gimmick" cap. No front panel sweep on the "D" unfortunately (HPIB only). So manual 10 kHz steps. Kiwi spectrum averaging and peak function to capture. Except for < 150 kHz where there is a clear drop-off, the scalloping is probably FFT spectral leakage since I never learned how to do my post-FFT DSP properly.
I see SMPS harmonics that are much louder than the fundamental all the time. Even into HF sometimes. It just baffles me. And when the fundamental is an unstable carrier, at HF it often turns into such a wide signal that the harmonics bridge together and it looks like a huge bump in the noise floor!
Wow, that was fast! Thanks so much.
I might be able to help with that FFT scalloping. Are you using a window ahead of the FFT?
I can see how SMPS harmonics might be shaped by the driving circuit (eg, push-pull vs single ended) but it is indeed hard to understand why they should be so much louder at higher frequencies. Maybe stray capacitance in (or around) the toroidal inductors in the filters? Poor quality capacitors?
And yeah, what might be a reasonably narrow line at the fundamental can get VERY broad at a high harmonic. Impossible to notch out.
Are you using a window ahead of the FFT?
It's using a Hanning window now. And the code has the usual suspects available (but commented out): Hamming, Blackman-Harris. I haven't look at the windowing specifically in years. Although we had the recent problems I induced when changing the FFT pixel downsampling and adding CIC filter compensation.
Do you overlap windows?
I'm a big fan of the Kaiser window for general purpose filtering, but I know that there are windows optimized specifically for spectral analysis (as opposed to filtering).
Addressing your noise issues and not the spectral display, I would encourage you not to immediately jump to whack-a-mole in seeking to improve Kiwi Performance. Rather, I have found it very productive to examine the coupling mechanism rather than simply trying to stay on top of suppressing interference sources.
For electrically small structures, which almost everything at LF and below is, the radiation resistance is minuscule. For virtually all situations we encounter actual inverse-square radiation from any sources/antennas is far below the interference levels the Kiwi reports. Thus I think it worthwhile to look at near-field and, particularly common mode (CM) coupling mechanisms.
In my experience, the dominant undesired coupling mechanism into the Kiwi is CM current over the path between wired LAN connection and the SMA-end of the Kiwi PCB. This includes the BB ground plane, cape connections and ground plane current paths on the Kiwi PCB. If one uses an isolated source having low self-capacitance (and then perhaps further reduces the potential for CM with a low inter-turn capacitance 1:1 transformer to create a test current source), -10 dBm on 15 MHz injected between the BB RJ45 shell and the Kiwi Antenna SMA results in about -85 dBm displayed on the Kiwi. This is more than 70 dB above the Kiwi noise floor in 1 Hz. At 100 kHz it's only down another 22 dB or so, still far above what the Kiwi can easily detect. It's for this reason that for each of my four Kiwis at the home QTH I have gone to WiFi interface, BBG/Kiwi's as described elsewhere on this forum and BBAIs with their native WiFi interface.
At LF, the noise floor of interest will depend upon the antenna system. Broadband electrically small antennas (they all are at LF and below) have antenna factors rising at 20 dB/decade while the ITU propagated noise, though all over the map with diurnal and seasonal variations, generally falls at about 25 dB/decade of frequency. Thus the noise limit of interest may be quite a bit above the Kiwi's native floor of ~ -157 dB/1-Hz but it still may be well below the kinds of levels that CM current injected via the LAN, PS or even GPS lines might be able to produce within the Kiwi.
For the Kiwi, and really every receive system we use "ground isn't ground" except by definition. But care in examining coupling mechanisms, reducing current in CM paths and symmetry (passive and active baluns) can really pay off and gains made here tend to apply no matter what new SMPS or other noise source pops up in the environs.
As a proof-of-performance it's also very useful to use symmetric antenna structures rather than single ended ones, e.g. monopoles, because the intended antenna, e.g. dipole, can be shorted to observe and confirm that the residue CM is not significantly limiting system performance. Considering the many-10s of dB of symmetry/CM_rejection we may need broadband RF baluns really are insufficient over the 3-4 decades the Kiwi covers.
When CM has been removed, there can still be the issue of near-field coupling to deal with, but I digress...
I've done some measurements on the frequency response of the Kiwi below 1 MHz. Using a -53 dBm tone (S9+20) from a calibrated signal generator, I got -53 dB on the S meter down to 50 kHz. It rolled off below that:
20 kHz: -1 dB
10 kHz: -3 dB
5 kHz: -7 dB
Hi Glenn, long time no hear! Wow, that's a really useful writeup that gives me a lot to chew on.
I'd been off HF for a long time but returned to it in my retirement. Physics hasn't changed, but everything else has. There's been an explosion in digital modes and signal processing that's great to see, and that's brought me back. But the flip side of computing is that virtually everything you plug into the AC supply grid these days has a switching power supply and/or is a nonlinear load of some kind, and the trash across VLF, LF and lower MF is just awful. There ought to be something I can do about it.
Desktop and server computer switching power supplies are of course the biggest offenders, mainly because of the amounts of power they handle. Sometimes there's room to install an integrated IEC receptacle/EMI filter. Even when you can, their small size doesn't permit much inductive reactance to low frequencies. Even the bigger EMI bricks lose a lot of attenuation < 500 kHz.
I agree, ground currents are a huge problem. I find this ironic. Wired digital communications has evolved from single-ended transmission over shared busses (e.g. the ISA bus; coaxial Ethernet) to differential transmission (often transformer isolated) over point-to-point links (modern TP Ethernet; PCI Express) precisely because grounding is so problematic at radio frequencies. Digital designers have learned to think like RF engineers, so how come so much radio work is still done with grounded, unbalanced lines that invite these problems?
I've connected my Ethernet switches in a fiber ring. Multimode fiber is surprisingly cheap and I really wish more computers natively supported it. Fiber is a rare new technology whose EMI benefits are not at all ambiguous.
But I still have a lot of short copper cables from those switches to my computers. Last year I replaced them all with Cat 7, which has shields around each signal pair and a shield around the whole cable between connector shells. I thought this was a great idea, and it may even be so at certain frequencies. But possibly not all. Maybe it's not such a great idea to bond all my equipment cabinets; seems like I'm just creating more opportunities for ground loops.
Interesting that you suggest WiFi to eliminate these problems. I've thought of that too. But isn't there something terribly ironic about using radio to avoid a radio interference problem? I'm a radio guy but wire still seems more reliable to me...
Do you overlap windows?
Not at the lower zoom levels because the waterfall/spectrum sampling is not continuous (the decimation factor is low and the sampling period is only a fraction of the display line time). This of course changes radically above zoom level 10 and I have to overlap when the sampling period exceeds the time it takes to display a line (the decimation factor is huge at that point). It took me forever to understand this because I didn't know what I was doing (some would say that has not changed, lol) and don't really have sufficient mathematical/DSP background.
Re Glenn's observations and experiments with CM and ground loops: I've been meaning to return to the topic and try and understand the issue in detail because Glenn is obviously a tremendous resource and understands a very important issue. But it needs a good chunk of time which I currently don't have.
The whole situation is confusing to me because of some anecdotal evidence. I don't deny it. I believe in CM. But I don't understand why some Ethernet-based installations have dead quiet noise floors and others are worse than useless. A lot of these quiet installations have photos of their setups in the Kiwi splash screen and they seem to be the typical jumble of equipment in a box with an Ethernet switch, SMPS, cables and other stuff all thrown together. And yet they work great!
What's the difference? Is there a minimization of the Ethernet/antenna cable loop by chance? Something subtle about the ground wiring/isolation? I don't get it. And I wish I did, because then I could make a recommendation to people.
Glenn, my configuration is too complicated to easily describe here, but I did try several simple experiments.
First, I disconnected the GPS antenna from the KiwiSDR. The noise < 30-40 kHz actually went UP. (This connector comes from an HP GPS antenna distribution amplifier shared with a NTP server, and a GPSDO powered from a linear 12V supply. It shares ground with my shack computer and a bunch of raspberry Pis in metal enclosures.)
With the GPS reconnected, I disconnected the antenna. All the noise disappeared except for a little at very low frequency. I touched the shell of the antenna cable to the shell of the SMA jack leaving the center pin open. No change. If there were common mode ground loops/currents through the KiwiSDR, shouldn't this have injected some noise?
I then reconnected everything and tried swapping Ethernet cables. I had been using Cat 7 cables with shield connecting the connector shells. Thinking that might help create a ground loop, I substituted a Cat 6 cable without shield. No obvious difference, though it wasn't easy to compare by eye since it took time for the Ethernet link to come back up, and the KiwiSDR doesn't always autonegotiate properly with my Linksys switch. (I could always take a screen shot, I suppose).
I've got a linear 5V supply supplying the Kiwisdr, a GPS distribution amplifier, and a Time Machine NTP server also using the common GPS antenna. Pulling power from the NTP server made no difference.
I have observed that Time Machine GPS stuff causes issues, apparently due to the plastic cases and the grounds. I bought one of the splitters, and had to set it aside as it caused (CMI?) noise issues.