The auto-wspr extension misses signals on a busy band
I don't think there is a fix for this, but I want to make the Kiwi community aware of what appears to be a limitation in the auto-wspr mode.
My Kiwi is running at a site with exceptionally good antennas and low ambient noise where the auto-wspr deamon frequently stops processing when it runs out of cpu time in the 2 minute window. At other times it misses many signals.
In my test one rx channel is set to auto-wspr on 20m while a second rx channel on the same Kiwi is opened from a Chrome browser session with its audio output fed through a virtual audio adapter to WSJT-x 1.8.0.
Here is an example of the difference between the 2 decodes on the same kiwi:
WSJT-x running on my 2011 Mac Air found 14 signals:
[rob@rob1 wspr]$ sort -k2,2rn spots.log
2330 15 6.8 14.097047 0 DM42LH 37 1260
2330 2 2.0 14.097181 0 K9YWO EN61 30 3083
2330 -2 1.7 14.097097 0 NI0L DM79 37 1569
2330 -9 1.6 14.097099 -1 KV4PD EM75 23 3379 <= Not found by auto-wspr
2330 -9 1.7 14.097186 0 W7REK DM42 27 1253 <= Not found by auto-wspr
2330 -10 5.7 14.097110 0 K8MDB EM79 33 3280
2330 -14 1.6 14.097131 0 AE2EA FN12 27 3893
2330 -17 6.8 14.097167 0 <K7RJ> DM42LH 37 1260 <= Not found by auto-wspr, a -32 dB sideband of <K7RJ> offset by 120 hz
2330 -18 1.8 14.097172 0 G0CCL JO02 37 8604 <= Not found by auto-wspr
2330 -19 1.2 14.097163 3 SA6BSS BP76 13 3457 <= Not found by auto-wspr
2330 -20 1.1 14.097152 -1 WB4CSD FM08 27 3811
2330 -21 1.6 14.097179 0 VE3XIX FN03 17 3717 <= Not found by auto-wspr
2330 -22 1.6 14.097087 2 W9RAN EN52 17 2911 <= Not found by auto-wspr
2330 -24 3.5 14.097124 0 N8MIQ EM79 23 3280
The auto-wsjtx found only 7 signals, all of which were also found by wsjt-x:
2018-05-22 23:30 K7RJ 14.097047 +18 0 DM42lh 5 KPH CM88mc 1258 304
2018-05-22 23:30 K9YWO 14.097180 +2 0 EN61fn 1 KPH CM88mc 3027 275
2018-05-22 23:30 NI0L 14.097097 -3 0 DM79mo 5 KPH CM88mc 1565 270
2018-05-22 23:30 K8MDB 14.097109 -10 0 EM79 2 KPH CM88mc 3265 280
2018-05-22 23:30 AE2EA 14.097130 -14 0 FN12fr 0.5 KPH CM88mc 3832 278
2018-05-22 23:30 WB4CSD 14.097152 -19 -1 FM08pp 0.5 KPH CM88mc 3820 283
2018-05-22 23:30 N8MIQ 14.097124 -22 0 EM79ti 0.2 KPH CM88mc 3324 280
Sorting by frequency suggests that perhaps auto-wspr does not discriminate as finely by frequency:
2330 15 6.8 14.097047 0 <K7RJ> DM42LH 37 1260
2330 -22 1.6 14.097087 2 W9RAN EN52 17 2911
2330 -2 1.7 14.097097 0 NI0L DM79 37 1569
2330 -9 1.6 14.097099 -1 KV4PD EM75 23 3379
2330 -10 5.7 14.097110 0 K8MDB EM79 33 3280
2330 -24 3.5 14.097124 0 N8MIQ EM79 23 3280
2330 -14 1.6 14.097131 0 AE2EA FN12 27 3893
2330 -20 1.1 14.097152 -1 WB4CSD FM08 27 3811
2330 -19 1.2 14.097163 3 SA6BSS BP76 13 3457
2330 -17 6.8 14.097167 0 DM42LH 37 1260
2330 -18 1.8 14.097172 0 G0CCL JO02 37 8604
2330 -21 1.6 14.097179 0 VE3XIX FN03 17 3717
2330 2 2.0 14.097181 0 K9YWO EN61 30 3083
2330 -9 1.7 14.097186 0 W7REK DM42 27 1253
2018-05-22 23:30 K7RJ 14.097047 +18 0 DM42lh 5 KPH CM88mc 1258 304
2018-05-22 23:30 NI0L 14.097097 -3 0 DM79mo 5 KPH CM88mc 1565 270
2018-05-22 23:30 K8MDB 14.097109 -10 0 EM79 2 KPH CM88mc 3265 280
2018-05-22 23:30 N8MIQ 14.097124 -22 0 EM79ti 0.2 KPH CM88mc 3324 280
2018-05-22 23:30 AE2EA 14.097130 -14 0 FN12fr 0.5 KPH CM88mc 3832 278
2018-05-22 23:30 WB4CSD 14.097152 -19 -1 FM08pp 0.5 KPH CM88mc 3820 283
2018-05-22 23:30 K9YWO 14.097180 +2 0 EN61fn 1 KPH CM88mc 3027 275
Rather than try to fix the Kiwi for this specialized application, I am trying to run multiple web browser sessions to kiwis and pipe the audio output of each session to an associated WSJT-x session on the same Mac/PC through a virtual audio connection.
My Kiwi is running at a site with exceptionally good antennas and low ambient noise where the auto-wspr deamon frequently stops processing when it runs out of cpu time in the 2 minute window. At other times it misses many signals.
In my test one rx channel is set to auto-wspr on 20m while a second rx channel on the same Kiwi is opened from a Chrome browser session with its audio output fed through a virtual audio adapter to WSJT-x 1.8.0.
Here is an example of the difference between the 2 decodes on the same kiwi:
WSJT-x running on my 2011 Mac Air found 14 signals:
[rob@rob1 wspr]$ sort -k2,2rn spots.log
2330 15 6.8 14.097047 0 DM42LH 37 1260
2330 2 2.0 14.097181 0 K9YWO EN61 30 3083
2330 -2 1.7 14.097097 0 NI0L DM79 37 1569
2330 -9 1.6 14.097099 -1 KV4PD EM75 23 3379 <= Not found by auto-wspr
2330 -9 1.7 14.097186 0 W7REK DM42 27 1253 <= Not found by auto-wspr
2330 -10 5.7 14.097110 0 K8MDB EM79 33 3280
2330 -14 1.6 14.097131 0 AE2EA FN12 27 3893
2330 -17 6.8 14.097167 0 <K7RJ> DM42LH 37 1260 <= Not found by auto-wspr, a -32 dB sideband of <K7RJ> offset by 120 hz
2330 -18 1.8 14.097172 0 G0CCL JO02 37 8604 <= Not found by auto-wspr
2330 -19 1.2 14.097163 3 SA6BSS BP76 13 3457 <= Not found by auto-wspr
2330 -20 1.1 14.097152 -1 WB4CSD FM08 27 3811
2330 -21 1.6 14.097179 0 VE3XIX FN03 17 3717 <= Not found by auto-wspr
2330 -22 1.6 14.097087 2 W9RAN EN52 17 2911 <= Not found by auto-wspr
2330 -24 3.5 14.097124 0 N8MIQ EM79 23 3280
The auto-wsjtx found only 7 signals, all of which were also found by wsjt-x:
2018-05-22 23:30 K7RJ 14.097047 +18 0 DM42lh 5 KPH CM88mc 1258 304
2018-05-22 23:30 K9YWO 14.097180 +2 0 EN61fn 1 KPH CM88mc 3027 275
2018-05-22 23:30 NI0L 14.097097 -3 0 DM79mo 5 KPH CM88mc 1565 270
2018-05-22 23:30 K8MDB 14.097109 -10 0 EM79 2 KPH CM88mc 3265 280
2018-05-22 23:30 AE2EA 14.097130 -14 0 FN12fr 0.5 KPH CM88mc 3832 278
2018-05-22 23:30 WB4CSD 14.097152 -19 -1 FM08pp 0.5 KPH CM88mc 3820 283
2018-05-22 23:30 N8MIQ 14.097124 -22 0 EM79ti 0.2 KPH CM88mc 3324 280
Sorting by frequency suggests that perhaps auto-wspr does not discriminate as finely by frequency:
2330 15 6.8 14.097047 0 <K7RJ> DM42LH 37 1260
2330 -22 1.6 14.097087 2 W9RAN EN52 17 2911
2330 -2 1.7 14.097097 0 NI0L DM79 37 1569
2330 -9 1.6 14.097099 -1 KV4PD EM75 23 3379
2330 -10 5.7 14.097110 0 K8MDB EM79 33 3280
2330 -24 3.5 14.097124 0 N8MIQ EM79 23 3280
2330 -14 1.6 14.097131 0 AE2EA FN12 27 3893
2330 -20 1.1 14.097152 -1 WB4CSD FM08 27 3811
2330 -19 1.2 14.097163 3 SA6BSS BP76 13 3457
2330 -17 6.8 14.097167 0 DM42LH 37 1260
2330 -18 1.8 14.097172 0 G0CCL JO02 37 8604
2330 -21 1.6 14.097179 0 VE3XIX FN03 17 3717
2330 2 2.0 14.097181 0 K9YWO EN61 30 3083
2330 -9 1.7 14.097186 0 W7REK DM42 27 1253
2018-05-22 23:30 K7RJ 14.097047 +18 0 DM42lh 5 KPH CM88mc 1258 304
2018-05-22 23:30 NI0L 14.097097 -3 0 DM79mo 5 KPH CM88mc 1565 270
2018-05-22 23:30 K8MDB 14.097109 -10 0 EM79 2 KPH CM88mc 3265 280
2018-05-22 23:30 N8MIQ 14.097124 -22 0 EM79ti 0.2 KPH CM88mc 3324 280
2018-05-22 23:30 AE2EA 14.097130 -14 0 FN12fr 0.5 KPH CM88mc 3832 278
2018-05-22 23:30 WB4CSD 14.097152 -19 -1 FM08pp 0.5 KPH CM88mc 3820 283
2018-05-22 23:30 K9YWO 14.097180 +2 0 EN61fn 1 KPH CM88mc 3027 275
Rather than try to fix the Kiwi for this specialized application, I am trying to run multiple web browser sessions to kiwis and pipe the audio output of each session to an associated WSJT-x session on the same Mac/PC through a virtual audio connection.
Comments
I have previously found that at KPH on 20M the Kiwi auto-wspr daemon runs out of CPU time in many 2 minute spot times before it can decode all of the spots. I'm not sure which version of the WSPR decoder SW is in the current Kiwi v1.195 SW, but I don't expect that the decode SW will be optimized for CPU performance anytime soon. ?
So at KPH I have turned of Kiwi 20M auto-wspr and been feeding Kiwi 20M audio to WSJT-x running on the Mac Air in the KPH rack. Since yesterday all of the KPH 20M spots have been recorded by that setup. Hopefully it will be a stable configuration, and I suspect I'll need to implement the same signal flow on 40M, 30M and perhaps other high WSPR traffic HF bands.
However I wondered if auto-wspr could be useful on bands with much less traffic. So last night I set up an audio feed from KPH to WSJT-x running on my Berkeley Mac which spotted as 'KPH/B'. That test showed some relatively small differences between the two spotters. Of the 293 unique spots logged by one or both of the systems, there were 59 spots which were only logged by one system, Of those 59 spots, 29 were unique to the auto-spotter and 32 were unique to WSJT-x.
Unfortunately I can't get a clean version of the comparison into this post
These results suggest to me that auto-wspr can be used on many bands at KPH. As we get more Kiwis at KPH I'll repeat this test on each band If I see little difference I'll use auto-wspr. I found this to be an encouraging result.