Signal fading is a constant issue for RTTY. Among the various methods to get around it is the use of frequency diversity. The signal fades when it takes two paths from the transmitter to the receiver, and the difference in path lengths is an odd multiple of a half wavelength. If the two signals arrive 180 desgrees out of phase and at nearly the same amplitude, a signal null occurs. However, a half wavelength at one frequency is a different distance at another frequency. So, we expect the multipath to form a comb filter with nulls at different frequencies. If we transmit our data redundantly on two different frequencies, there is a chance that when one frequency is in a null, the other is not.
Amateur RTTY has largely standardized on a frequency shift of 170 Hz. Originally, the shift was 850 Hz. An experiment was run to determine how frequencies faded when separated by 170 Hz and 850 Hz.
A ten minute audio file with the following simultaneous tone frequencies:
The resulting CSV data is graphed.
You can zoom to an area of the graph by left click dragging over the area. Right clicking takes you back to the full graph.
Because the recording was played back twice and the start of the MsLevel command was executed some variable amount of time before the playback of the recording started, times in one graph are not exactly the same as times in the other graph.
Looking at the 170 Hz graph, we see a deep null at 203 seconds. Zooming in to the null, we see the two signals hit minimum about 14 ms apart. In addition, the signal levels drop from about 0.12 to 0.001, a 42 dB drop in signal level.
Looking at the 850 Hz graph, we again see substantial fading. Zooming in to the 203 second area, we find the minimum levels of the two signals are separated by about 180 ms. Still, there are very deep fades affecting both the mark and space frequencies. There may be a slight frequency diversity advantage to 850 Hz shift, but it appears minor.
A wide variation in received signal level is not bad since it can be corrected by an AGC or limiter. However, if the signal fades into the noise, data recovery is not generally possible. If the two reflected signals have the same path attenuation, increasing transmit power does not remove the null since both the paths result in the same higher signal getting to the receiver. However, since the path lengths vary over time, a higher power can result in a shorter period of complete null. Further, if the paths have different attenuations, a null is not a complete cancellation of the signal. Higher transmit power could make this incomplete null have a signal level above the noise allowing data recovery.
It appears that use of 850 Hz shift results in a slight improvement due to frequency diversity.