Here is some "Food for Thought" -- some ideas I recently developed for improved SSTV sync detection that I don't believe have been addressed in the past, and might be worthy of discussion and/or consideration.
My TriplePIC scan converter employs 1200 Hz bandpass filters, full-wave detection, and post-detection low-pass filtering that aim to achieve "matched filter" performance. Despite these efforts, some minor line-to-line jitter can still occur under certain circumstances. I recently looked into this issue to see if any design improvements could be made to further reduce the instability.
It should come as no surprise that some of the basic assumptions we have made about sync detection are less than ideal. We assume that sync, and sync alone, occupies 1200 Hz.
It doesn't. There is ingress by noise, interference, and video sideband energy that extends well below 1500 Hz. These effects modulate the amplitude of the raised-cosine pulses coming from the horizontal sync low-pass filter.
The threshold assigned for the voltage comparator that follows the LPF is set to a fixed level (ideally at half the peak amplitude of the pulse) because we think that amplitude matters.
It doesn't. Noise causes the pulse produced by the comparator to become pulse-width-modulated. The horizontal sweep gets slightly retarded and advanced as a result, which leads to the mis-registration problem.
Sync is the clock that drives the whole system. What we need is circuitry that is sensitive to
sync timing, not amplitude.
So, instead of setting a comparator threshold somewhere on the rising or falling edge of the raised-cosine pulse where it is sensitive to pulse amplitude, how about detecting
the time where the pulse reaches it's peak amplitude, regardless of what that peak voltage may be?
One of the easiest implementations might be to take the time derivative (dV/dt) of the raised-cosine pulse. dV/dt will rapidly transition through zero volts when the filtered sync pulse (plus noise) reaches its maximum amplitude, whatever that specific amplitude may be:
- raised_cosine_derivative2.png (45.3 KiB) Viewed 9988 times
I admit that I have never seen or heard of this approach before, but tests I have conducted thus far show essentially jitter-free differentiator zero crossings under conditions where a tried-and-true comparator-sliced sync pulse shows instability.
More news to come. In the meantime, I welcome comments, and offer the following supporting evidence.
Here is a spectrogram plot of PA0KLS's Testbeeld image as an 8.5 second, 128 line, 60 Hz SSTV image. The SSTV video was generated using software of my own design, using 16 bits and a sampling rate of 44.1 kHz:
Next, in order to illustrate how much video sideband energy falls across the sync frequency, I altered my SSTV generation software to move the 1200 Hz sync pulses to 1750 Hz, placing them in the nominal center of the SSTV passband. The black lines illustrate the 3 dB sync detection bandwidth used in my scan converter. It is easy to see what parts of the image where the "QRM" and the corresponding sync instability occurs:
A "real-world" image containing fewer high-frequency components. It almost looks as though there is an intentional guard-band around the sync frequency, even though there isn't:
Finally, for clarity, a simple ramp from black to white in 8.5 seconds:
Remember, don't fault the image contents for having video sideband energy that clobbers the sync frequency. We need to be able to tolerate that just as much as noise and QRM would do the same over-the-air.
73 de John, KD2BD