OwenDuffy.net 

This article reports an experiment to compare two adjacent antenna systems on 40m over a 14,600km path using QRSS.
The experiment is to compare a temporary 40m quarter wave vertical with a more permanent structure with more radials at greater height.
The receiving station was W4HBK at Pensacola, FL (USA).
The transmitting station was VK2DVK located about 100km S/SW of Sydney (Australia), and the antennas were two adjacent quarter wave verticals over elevated ground planes. The two antenna systems are:
At the time of the tests, the path elevation angle at the transmitter is very low, in the range 2° to 6°, and bearing is 78°.
A QRSS keyer that facilitates antenna switching based on special characters embedded in the message was constructed. It was used with a quite standard Icom IC7410 in CW mode adjusted for 5W output.
Fig 1 shows the internals of the QRSS keyer. The keyer is described in detail at Another Morse beacon keyer  A/B RF switching.
The message is structured to send a period of keydown for 20s, then VK2DVK at QRSS6. Antenna B is used to send DVK, the rest of the message is sent on Antenna A.
The receiver used Spectrum Lab to gather and present a view of received signal and noise.
Two kinds of charts are used for analysis:
Data is presented below for the hour from 08:00UTC on 05/01/13.
Fig 2 shows a set of six graphics cropped from the full waterfall charts. They show VK2DVK's signal at the top of each chart over the hour.
The watch plots below plot the calculated Signal/Noise ratio in blue, and this is the best indicator of the relative performance of each transmitting antenna at the time. The green and red lines are Signal and Noise respectively.
Fig 3 shows a set of six graphics cropped from the full watch plots. They show VK2DVK's signal over the hour.
A frequency analysis was performed on the watch plot data to find a S/N figure that divides the data into signal on / signal off periods.
Fig 3 shows the frequency distribution of 6348 observations during the study hour. It is clearly bimodal as a result of the signal on / signal off periods and the threshold 5dB will be used to separate the observations that probably relate to the signal on condition. This method is not the best, but the underlying problem is the modulation and removing that is the key to better statistics.
Next, the observations are divided into those in the first 320s of each 10min period when Antenna A was used, and the rest when Antenna B was used.
Fig 4 shows the distribution of observations for Antenna A and Antenna B. The high observation at the low end of Antenna A hints that some signal off condition observations have been captured which will drag the mean for Antenna A down a little, and truncating the data distorts the distribution a little.
The S/N is taken to be a log normal distribution (Duffy 2012), so S/N in dB a normal distribution.
Mean  SD  n  
Antenna A  9.3  2.6  1736 
Antenna B  11.6  3.1  1294 
Table 1 shows the summary statistics for the observations attributed to Antenna A and Antenna B. With such a large number of observations, the difference in SD of each of the sample sets is larger that might be expected.
Difference in means  2.3 
Grand SD  3.84 
df  3028 
t  21.8 
p  3e98 
Table 2 shows Student's ttest applied to unequal sample sizes and equal population variance. The t statistic is very high for a very large sample set and the probability that the means are different by 2.3dB by chance alone is 3e98, extremely small!
Lets not get too carried away with the ttest result, the method of apportioning observations to signal on and signal off is a bit rough, and the message content for each antenna is not identical so the measurements are exposed to some error due to detector response behaviour.
The first question is whether one antenna is significantly better than the other?
By eye, it does appear that Antenna B might have an advantage of perhaps 3dB over Antenna A.
Concentrating on the S/N plots when a dah element is being sent, on some of the 10min plots, it is difficult visually to assert that one antenna is better than the other. There do seem to be more instances when Antenna B is poorer than Antenna A, but without some statistical analysis of the results, the difference would seem to be less than 2dB and quite small in terms of the variation in S/N ratio due to normal propagation fading.
Height of radials  Gain (dBi)  
Radials  
(m)  (λ)  3  4  8  32 
0.005  0.000119  4.7  3.6  1.6  0.3 
1  0.023  0.2  0.2  0.1  0.1 
3.5  0.083  0.22  0.12  0.1  0.13 
Table 3 shows the results of an NEC4 model of the maximum gain of a 40m quarter wave with 3, 4, 8, and 32 radials at 0.005m, 1m and 3.5m above average ground.
It can be seen that NEC4 would suggest there will not be much difference between an antenna with 4 radials and another with 8 radials at 3.5m, yet the plots in Fig 3 suggest that there might an advantage of perhaps a few dB in this test. The fading cycles are a large source of statistical noise, and some further study of captured data might give a more accurate answer.
Statistical techniques can be used to mine the observation data, and in this case indicated that the Antenna A was 2.3dB better with an extremely high significance, but the underlying test with different message content used for each antenna, and the method of dividing the observations between signal on and signal off condition are exposed to errors. A better test is to use long periods of unmodulated signal and select the signal on / signal off condition based on time. The latter calls for tight syncronisation of transmitter and receiver which proves challenging in practice.
Version  Date  Description 
1.01  05/01/2013  Initial. 
1.02  
1.03 
© Copyright: Owen Duffy 1995, 2021. All rights reserved. Disclaimer.