Senstivity comparison to Airspy HF+, trying to understand the data
Apologies in advance since this is my first KiwiSDR, and I am NOT an EE. I am trying to understand why it seems significantly less sensitive than the Airspy HF+ SDR I have. Aside from what my ears were telling me, I decided to make some objective measurements to share, on the chance someone can help me understand what I am doing wrong. Or guide me on some better experiments to verify that my KiwiSDR does not have a hardware issue.
With no antenna connected, AM mode (at 10 kHz bandwidth) the 0-30 MHz spectrum is reasonably clean, with a typical reading of -116 dBm. If I zoom in to a narrow slice, say 30 kHz wide, and enable spectrum display, the visual noise floor level is about -153 dBm. This makes sense to me since the power bandwidth for the spectrum display will be much smaller than for the 10 kHz detector bandwidth. Also, since I am using a switching wall wart power supply (shame on me, but it's all I have available for now), I also powered the Kiwi via an external USB battery....with no discernible impact on the noise floor. Question #1 - are these observed values typical? They seem OK to me, so far so good. Or is it?
Next, I attached a -126 dBm signal source (TinySA, low output, attenuated from -76 dBm to -126 dBm, unmodulated carrier at 11.500 MHz). While the dBm the Kiwi reports doesn't budge from -116 dBm, the zoomed spectrum display does show a peak about 6 dB above the noise floor. It's noisy, but enough signal to hear in CW mode, or USB mode if I lower the tuned frequency to 11.4995 MHz (500 Hz offset). in sharp contrast, with the Airspy HF at this level the signal sounds much less noisy, and the S/N ratio (displayed in SDR# is anywhere from 16-34 dB, depending on the sampling rate I set, i.e. at the highest 912 kbps rate the S/N is 16, increasing to about 34 dB at a 24 kbps sampling rate. Why the difference versus sampling rate is a separate mystery, but regardless of that, even at the lowest S/N I can hear a large difference versus the KiwiSDR. I have other data collected for higher signal levels (up to -76 dBm), but the -126 dBm level illustrates the pattern I am seeing. Question #2 - is the KiwiSDR response to a -126 dBm input signal typical?
My next experiment will be attach the outdoor antenna to both SDRs and pipe the audio (virtual audio cables) into two separate instances of WSJT-x running on a Windows PC, and see how the WSPR decode counts compare. That seems like another way to compare the weak signal performance.
Open to any and all suggestions and/or editorial comments!
Yeah, so this is a huge can of worms as they say. The subject of receiver noise, and SDR noise in particular, is a mixture of generic noise topics (that have been around forever), SDR-specific issues (e.g. ADC, ADC frontend, SMPS) and Kiwi-specific issues (CM noise on cabling, Ethernet etc).
There are people on here that can address these issues much better than I can. But I will say a few things. First, you absolutely have to get rid of the SMPS feeding the Kiwi. And Any other SMPSs that show up in the VLF/LF spectrum of the Kiwi. It has been well documented that the harmonics of some of them extend into HF and can be quite strong. Worse, because the fundamental carriers are almost always unstable, by the time they are multiplied to HF they spread out and look like a broadband elevation of the noise floor. Similar in some ways to what VDSL can do.
Just go look carefully at the harmonics of an SMPS in the Kiwi spectrum/waterfall starting at the fundamental going up and you'll probably see what I mean. So many Kiwis on rx.kiwisdr.com suffer from this it's just terrible. OTOH, there are lots of Kiwis there with excellent S/N. So it's not a flaw in the basic archeciture (yeah, the ADC frontend could use some improvement I've heard).
Secondly, the Airspy HF+ does have better sensitivity than the Kiwi because it is a totally different architecture. It uses a 24-bit delta-sigma IF (not RF!) ADC downstream of a polyphase mixer (a fancy name for a more-than-4-phase Tayloe detector). BUT you pay for this by the HF+ having limited sampling bandwidth. You cannot get a full 0-30 MHz spectrum/waterfall out of an HF+ like you can on a Kiwi. Or even half that. That's the tradeoff. No free lunch!
Thank you for the detailed response to my query. I'm still finding my way with this SDR, and yes indeed the 30 MHz bandwidth really shines compared to other SDRs. I am almost relieved to hear that the Airspy HF may be more sensitive than the Kiwi, suggesting my hardware is "normal". Given the levels of atmospheric noise, the raw sensitivity on HF is not the be all and end all, of course.
That said, I am hoping there might be a KiwiSDR user out there that could feed a -126 dBm unmodulated carrier to the input connector at 11.5 MHz, and report what they see in terms of the spectrum/dBm reading etc.
Thankfully the SMPS I am using shows no signs of bleeding into the waterfall, at least with no antenna connected for test signals. Same with an external USB battery connected. However, with an antenna connected there are some tell-tale harmonic spikes that emerge, either from common mode coupling to the feedline or transfer into the house mains etc etc. So long term yes a linear power supply is in the works.
Remember also that the waterfall/spectrum and S-meter have calibration factors on the admin page, config tab. Neither the waterfall nor S-meter magically measure dBm by themselves. They measure relative dB and the cal establishes the reference. The defaults are -13 dB which is what I got by feeding an hp 8657b directly into the SMA input and trying a bunch of frequencies and levels. Others recommend slightly different values.
The response of your antenna of course throws all of that out the window. But you can make relative comparisons using the same antenna between different receivers with some assumptions about impedance matching etc.
In our experience, at many quiet QTHs like KPH and KFS, the Kiwi's sensitivity is limited by its internal noise figure above 10 Mhz. With 15 dB of gain in an LNA and a high pass pre-emphasis filter which attenuates 10 dB below 10 MHz in front of the Kiwi's RF input, we have matched the performance of high end receivers like the Flex.
Of course the 15 dB gain LNA compounds the overload problems from AM broadcast signals, and with the increase in sunspots overload from MW broadcast signals will become a greater source of overloads. The Nooelec AM blocking filters work well for most sites, but blocking the MW broadcast signals on the 41M Broadcast band can be a real challenge.
I did some additional comparisons for WSPR decoding yesterday with the same antenna and audio piped into separate instances of WSJT-x via virtual audio cables. The spot count was consistently higher for the Airspy HF+. Looking at the signal reports for the spots, on 20 meters the Kiwi was about 7 dB down, and 2 dB down for 40 meters. But I don't know if that is a fair comparison, i.e. is the KiwiSDR audio from the browser interface of the same nature as what wsprdaemon would be working with?
I also compared the response of the KiwiSDR to the (humble) RTL-SDR dongle using an unmodulated 11.5 MHz carrier from -126 dBm to -100 dBm. The S/N ratios were generally in the same ball park, and matched what my ears detected in CW mode. Of course the dongle has a slew of other limitations, but in a pristine signal environment the sensitivity relative to the KiwiSDR surprised me a bit.
I'm gradually beginning to comprehend what a can of worms I've opened with this comparison testing, given all of the variables and pitfalls. But I am learning a lot and having a lot of fun in the process.
the audio for wsprdaemon stays digital the whole way. A WAV file is recorded and decoded by wsprd, the decocer that is p/o wsjt-x.
On the Kiwi, be sure to turn off the audio compression when using a browser to feed audio to WSJT-x. In the default 'COMP' audio mode, the Kiwi compresses the audio before sending it over IP to the user's browser. Uncompressed audio sounds better and in my experience delivers more spots and better SNRs.
It's easy to make non-useful comparisons. I would like to add that Rob's comment about limiting noise floor at a quiet site presumes a matched antenna. This allows comparison with ITU noise floor values. As he implies, dynamic range is perhaps a more important factor than sensitivity at a particular point with a particular antenna mismatch.
A 'normal' Kiwi generally shows about -157 dBm/1-Hz noise spectral density and overload at about -15 dBm (instantaneous) aggregate signal level. In the presence of a large input that tickles the MSB of the ADC it may be 2-3 dB worse than this due to ADC non-linearity.
This means that one can get within 18-20 dB of thermal noise which at a quiet location may occur somewhere in mid-HF on a well matched antenna. But if there is mismatch to the radiation resistance of the antenna, the required performance to avoid receiver limitation of incoming SNR is more severe. A 1m dipole and a half-wave dipole on 160m each have essentially the same SNR but voltage levels and radiation resistance of the smaller antenna are much smaller - forcing one to use low noise gain ahead of the receiver, while keeping other signals that might be impinging from driving the system into overload. That's the dynamic range story. Low noise gain is cheap and easy (at 50 ohms, anyway!) but high dynamic range that can be well-coupled to a broadband antenna system is another matter entirely.
BTW, on a fair sample of Kiwis, I've found that an S-meter and WF setting of -16 to generally give accurate results at mid-band, 15 MHz. There is a bit of roll off on the low side and typically a couple of dB on the high side but this is a reasonable all-round value. To have it be meaningful though, one must know the gain/loss from the radiation resistance of the antenna system being used to that Kiwi SMA connector.
Hope this helps.
As always, thank you for your patience in explaining the facts about noise levels. For my own curiosity I made some signal level measurements with the KiwiSDR, using a TinySA spectrum analyzer as the source. The TinySA is not a precision source, but has been reported to be within a dB or so of true values. Overall I was surprised at how close the values were in agreement, given all of the potential sources of error (including the uncalibrated 50 dB attenutaor I needed to use to reach as low as 126 dBm). Based on what I have read elsewhere (KA7OEI blog) these results give me assurance that my KiwiSDR is performing as it should.
dBm output from Tiny SA* dBm reading on KiwiSDR (10 kHz bandwidth, AM mode)
-126 -116 (-147)**
-120 -115 (-142)
-110 -111 (-113)
-100 -102 (-112)
-90 -91 (-102)
-80 -80 (-95)
-70 -70 (-84)
-60 -59 (-73)
-57 -55 (-70)
*50 dB attenuator in place to yield -57 to -126dBm output range, 11.5 MHz unmodulated carrier
** values in parentheses are my estimated peak readings from the zoomed spectrum display
Use the S-meter/dBm display not the spectrum or waterfall to measure. You need the 24 bit path not 16. Using spectrum/WF for absolute levels is risky on multiple counts. One has to keep in mind the bin width, that the signal actually landed within the 'maximum spot' for the particular zoom level. On to ;o fof that ther is the peak/average noise problem and averaging.
The S-meter seems to do a credible job on both coherent signals and RMS noise. This is not the case for the spectrum/WF.
In fact, in general I'd caution against using the the spectrum/WF for anything other than relative measurements - and not even the WF for these since colors aren't a very good way to distinguish.
Yes, I agree only the S-meter/dBm is usable for true values. The odd formatting on my post with all the numbers made it a bit ambiguous. For each line the first value is the power output from the TinySA, the second value is what the S-meter/dBm display read, and the third value (in parentheses) is my "guesstimate" from reading the spectrum display peak. All in all I'm happy to get such good agreement.
-126 -116 (-147)
-120 -115 (-142)
-110 -111 (-113)
-100 -102 (-112)
-90 -91 (-102)
-80 -80 (-95)
-70 -70 (-84)
-60 -59 (-73)
-57 -55 (-70)
Also, be aware that the S-meter 'hits the pin' and won't display values lower than -127 dBm, thus it is necessary to assure enough noise bandwidth to generate a larger reading, A 1 kHz BW though it might help dodge coherent signals may not be high enough to measure the Kiwi's own floor.