On 19/05/2013 00:58, Hans Summers wrote:
Many thanks for all your efforts in helping to collect evidence to track down this problem. I know it might be a little frustrating but thanks for persisting!
I suspect that there might be some arithmetic issue in my code which reduces the accuracy of the calculation when the frequency is high. More specifically, when the shift frequency becomes smaller relative to the baseline output frequency.
If you have time, I think that a very useful test might be a ramp from 0 to 5,5Hz shift (say), at 0.5Hz increments. A staircase, really. This would show very clearly any issues with inaccuracies of the steps, and clarify immediately on what bands and frequencies they started to become a problem.
I am referring to the stuff in section 8 of the manual (page 28). A customised message to produce a staircase of shifts from 0 to 5.5 Hz, with 5 seconds spent on each stair. The staircase will take 1 minute. This will make it much easier to see what is going on than images of WSPR, which spends only a short time on each frequency so looks messy on the Argo (etc) display.
The required mode should be FSK/CW, and the Message string contains the speed and the shifts, with an asterix ' * ' character to start and end it. So it should be *050123456789AB* for the described ramp.
The efforts are perverse fun Hans, as long as the information flow is not too great for you.
I will do the ramp test later today. We are thinking along the same lines. What I have done this morning is to set FSKCW5 to a 9Hz (Max) shift and run that at the 28MHz WSPR frequency to see if the system is capable of shifting the 6Hz for WSPR or more. This has proved that it is indeed capable and pretty accurate.
Thinking also on the generation of WSPR and the V2 problem with the System Clock I wondered if there may be an accuracy problem effecting the arithmetic there. Since the GPS works fine for me I have never worried about setting the System clock. I checked and it was 1.740KHz high, I calibrated accordingly, set GPS to off for good measure, set the time manually and ran WSPR at 7MHz all looked well and it decoded as usual. I then tried 28MHz WSPR and the shift is too small and there are no decodes.
So that is another pathway eliminated.
BTW consistent with FTD's intermittent shift on 28MHz WSPR it is difficult to say but it may vary over a range over time, I am pretty sure the excessive shift on 7MHz went from 10Hz to 8Hz after a few hours last night.
As for Digi WSPR looking messy it is a pity that DDS and WSPRaspberryPi systems can't do a soft slide of frequency shift. Having heard a semi local station yesterday on 28MHz clicking and banging away on WSPR I must say that the incessant spikes are disturbing, even anti social.
Does the U2 DDS just change the frequency word for each Data transition, it sounds as if it is going to zero in between. I think that the really bad noise, (hammer on a spade) on 28MHz WSPR though is another manifestation of the bad shift, the tones hitting the ceiling at Data3.
73 Eddie
