Antenna A/B testing – part 2, enlist AI

There’s a new technology in town. Let’s persuade ChatGPT or GROK to do the grunge work. That’s what robots are for. Right?!

Lord Callum McCormick demonstrated how he uses AI to do the tedious work and to produce wonderful comparison charts. Basically, he asked ChatGPT to separate results stored in a spreadsheet and do various calculations on the two sets of results. Callum was seeking which antenna is “better.” But, better at what? Well, that’s why we want various calculations; number of stations that heard us, number of times they heard us, distance, and signal strength, etc.

Callum used ChatGPT. I stole his prompt and did the same with Grok. It’s fascinating watching the AI machine, collect the information, learn about how WSPR works, build a program to separate various data and do calculations. It tells you what it is doing as it goes along, and then shows the results you asked for. Be aware: one has to be precise in what we ask for. Some report that AI can sometimes give you more than you expect. Adherents of the AI industry describe these adventures into uncharted, unspecified, territory as “hallucinations.” My antennas are mysterious enough without hallucinations. Just give me only what I asked for. Thank you.

Callum specifies the prompt he used in his video. I copied it and modified it a bit for my own interests, omitting his RX analysis. I simplified the process to need only 2 log file uploads. We arrive at very similar results. Since I use most of my antennas enough to have a feel for them, these tests confirm and solidify what I know about them. They also give me a quick way to compare a newly built antenna with one of my long time favorites. It quickly eliminates the “Is it the antenna, or today’s propagation?” question.

Process:

• Set up both antennas
• Connect them to an A/B switch
• Set up the WSPR transmitter and use a PC to control the transmitter. Set the callsign, grid square, and band for the test.
• Start running.

• Watch the WSPR Transmitter control screen and every time the little red light shuts off, switch the A/B switch. Missing this will spoil the results.
• Continue for as long as you please. I like to go through 6-8 cycles for each antenna
• Collect the results from wsprnet.org. My preference is to sweep the data into a spreadsheet. Name that spreadsheet Transmit log.xlsx
• Create another spreadsheet named Time slots.xlst

• Fetch the prompt from the bottom of this post. NOTE WELL: Edit at least the line which specifies the label for the report.
• Upload the A/B prompt and the two spreadsheets to Grok.
• Sit back and watch it work.

Below is part of what Grok reported. I’ll explain this particular test in the next post. Or, you can go here to see all of what Grok prepared. Note that at the top of the Grok report you can see the actual prompt I used (feel free to steal for yourself) and the data files. Next post: more detail about this particular comparison.

72 de N4REE, Bob

===== Below is PART of the results from Grok =====

Overall Statistics

Metric	POTAPerformer (PERFormer)	EFHW
Total number of reports	384	236
Number of unique reporters	128	95
Median SNR	-17	-18
Median distance (km)	1551	1498
Maximum distance (km)	6599	3949

Distance Bucket Analysis

Bucket	POTAPerformer (PERFormer) Count / Median SNR	EFHW Count / Median SNR
Short haul (0–2000 km)	321 / -15	204 / -17
Medium haul (2000–4000 km)	62 / -18	32 / -19.5
Long haul (>4000 km)	1 / -26	0 / null

Observations on Antenna Behavior

Based strictly on the provided data, the POTAPerformer (PERFormer) generated 63% more total reports and engaged 35% more unique reporters than the EFHW, indicating broader overall reception during the test periods. It also achieved a slightly higher median SNR across all distances (-17 vs. -18) and a marginally higher median distance (1551 km vs. 1498 km), suggesting more consistent signal strength and reach under the tested conditions.

In short-haul distances, the POTAPerformer received…. (See more at the GROK report link)

---

Grok A/B Prompt

I have conducted a controlled A/B antenna comparison using WSPR only.
Both antennas were tested: at the same station on the same band at the same power within clearly defined time windows.
Nothing else was changed.
Find two files attached.
Both files are in Excel format.
1 The Timeslots Log in UTC notes which antenna was in use and exactly what time.
2 The Transmit log downloaded from WSPRnet in UTC.
The logs are accurate and in clear two-minute chunks – there is no need to be concerned about the times. 
My timeslot log includes the antenna label for every 2-minute WSPR slot. 

Your tasks: 
Label this report "A/B EFHW vs. POTAPerformer - 20m"
Identify which antenna was in use for each transmit period.
For each antenna, calculate:
- total number of reports
- number of unique reporters
- median SNR (not maximum)
- median distance
- maximum distance

Group TX results into distance buckets:
- short haul (0–2000 km)
- medium haul (2000–4000 km)
- long haul ([greater than] 4000 km)

For each bucket, show report count and median SNR. 
For each antenna, calculate:
- total decode count
- median distance (if available)
- maximum distance

Output requirements: Present results in clear tables.
Do not declare a single winner.
Explain what each antenna appears to be better suited for - based on the data.
Focus on differences in behavior, coverage, and application rather than preference or opinion.
Base conclusions strictly on the data provided, not on predicted behavior.
Explain how you calculated the results.