From the CHANGE_LOG file:
v1.455,456 May 19, 2021
Config tab: Added button to reset passband fields to default values.
Connect tab: Improved proxy setup UI.
Right-click menu: Added "snap to nearest" mode. Setting is remembered/restored per Kiwi
visited just like freq/mode etc. Waterfall cursor differs when snap mode active.
When snap is active the freq step keys (j/i, left/right arrow keys) are modified so
the ctrl/alt-key steps the smallest amount and the unmodified key steps the same
amount as the snap (e.g. 9/10 kHz on MW). Usually the unmodified key step the
Changes to more easily support general (even non-SDR) FPGA development.
v1.457,458 May 21, 2021
Waterfall / spectrum:
Fixed low-level post-FFT artifacts spotted by Marco's SuperSDR app. Thanks Marco!
Added CIC filter compensation to better level spectrum at top end.
Added GM-11F Sevastopol 7090.9 kHz per fabrys on Kiwi forum. Thanks fabrys!
This update has certainly resulted in some changes to the FFT output!
Quite obvious on the spectrogram. I expect it will also result in a change in SNR metrics.
Make sure you didn't end up with 457. Restart to get 458. 457 had a terrible bug and there was a small window of time (like 15 minutes) when it was available.
Actually, this spectrogram must have been made by sampling the waterfall at zoom=0, right? The bug that was fixed had to do with interpolation problems at higher zoom levels, when averaging, causing visible artifacts.
Because zoom=0 is a special case that is handled differently in the code (because e.g. there is no CIC decimation in the FPGA) it's possible the bug fix is causing a problem with the peak values when zoom=0 as your spectrogram shows.
I will look into this..
Everything looks good here, even with zoom=0. In fact all of the waterfall images look slightly better after the fix (less noisy).
For each of your time sample points are you taking a bunch of consecutive waterfall measurements and averaging them? I can't remember if you're using kiwiclient or accessing the API on your own.
Yes, it averages over 100 samples:
Edit: The kiwi is on 1.458.
It seems that the waterfall amplitude (on the kiwi webinterface) is now slightly higher on the high frequencies, and lower on the low frequencies, just like in vk3qi's diagram.
Spectrum with antenna disconnected:
@HB9TMC I'll fix that today. Shouldn't be applying the CIC compensation when there is no CIC in use! 🙄
As for the spectrograms, I don't know. All the Kiwi spectrums I looked at using the Kiwi spectrum display and high averaging gave the same peak values on a variety of settings. I guess the only thing to do is make a new waterfall API option so you can select either behavior and Marco can get his fix for SuperSDR.
Thanks John, yes I noticed same increasing gain slope here on zoom=0.
Looking forward to the fix.
Short version: Old behavior restored so external apps, including those using kiwiclient et al, should work as before. But Kiwi web interface uses new changes as described below.
v1.459 May 23, 2021
Waterfall / spectrum:
Fixed the CIC compensation added to the last release.
Added waterfall interpolation URL parameter "wfi=N" e.g. kiwi:8073/?spec&wfi=13
Also available as an API parameter on the waterfall web socket: "SET interp=N"
Value N below. Add 10 to enable the CIC filter compensation.
The value below answers the question: "What to do when multiple FFT values
contribute to a single waterfall/spectrum pixel bin". In most cases the FFT
is larger than the 1024 display pixels of the waterfall (or external application).
0 take max (old and continuing server default for kiwiclient, external apps etc.)
1 take min
2 take last
3 do drop sampling
4 take cumulative moving average of values
13 drop sampling + CIC compensation (new Kiwi browser interface default)