Retro SSTV anyone?

[Klaas Robers]

Nice picture Steve. It looks as if the frequencies are now much better.

If there would be "spurious" creeping into the VCO then this should be visible on the monitor screen, but even more on the oscillogram of the video wave form. I see nothing of that.

But what me also surprises is that the "spectrum" is roughly cut off at the upper edge. Wouldn't that graph go higher up? And how would the peak form be there? Is this the only setting for these graphs? Would it be normal that the graph is flat topped?

After better observing it looks that my sync is now slightly too high. I will have a more precise setting later in time. On the other hand, a VCO is always a somewhat unstable oscillator. On my oscilloscope the video line of a black picture is razor sharp, as is it on a white picture. Spurious in the VCO input would then be visible.

[Steve Anderson]

I think the spectrum display in MMSSTV is to be thought of more as a tuning aid or tuning indicator when receiving via SSB, not as one would think of a real spectrum analyser. A guide if you will. It's accurate enough in frequency when fed a fixed tone but as for amplitude - well, anyone's guess.

I made up a chequerboard file similar to your own with eight squares rather than six, 256 doesn't divide by six easily. It's also easy to see where it came from.

The screen-shot, spectra and waveforms are below. The waveforms are the baseband input to the modulator (magenta) and the demodulator output (yellow), The overshoot is due to the demodulator output filter which I'm still musing over.

I'll get on with Testbeeld3 right now.

Steve A.

[Steve Anderson]

Testbeeld3, screen-shot and waveforms below, I wasn't able to get a spectra for it, I think the Windoze PC needs the usual early evening re-boot. Well you can't expect the thing to run for more than six hours without some limb falling off...

You'll notice the sync detect output (cyan trace) is late, this is the result of making sure that it really is syncs. Seven samples in a row need to be below the set threshold before sync is acknowledged, this stops runt pulses falsely triggering, the actual trigger comes from the trailing/rising edge of the sync pulse. Non-standard I know, but it works.

Also the 1200Hz grating is of a greater amplitude.

Ignore the usual crud top and bottom of the screen-shot.

The other screen-shot is of my local when I'm in the UK. Ye Olde Kings Head, Church Street, Epsom, Surrey. A lovely little hole-in-the-wall.

Steve A.

One thing that might be 'educational' is to do a baseband to baseband screen capture and see how much the mod-demod process impairs the picture...

[Klaas Robers]

Steve, nice pictures and oscillograms. I see that your synchronisation improved a lot. It is also good to see that the small arrows on the Testbeeld3 are fully visible. And I see too that the 1200 Hz has a larger amplitude. It is comparable to what I see on my oscilloscope. I think it very difficult to make a photo of those particular lines as the exposure time of my camera is at least several lines, so I have to take it on the guess.

What strikes me is that your demodulated and filtered oscillograms show no ringing at the towards white going edges. They show ringing (overshoot) on the down going edges. That is normal and cannot be avoided if you want to have a rather steep cut off. Only in Bessel filters there is no ringing, but they almost don't filter.

With an allpass you can spread the ringing to 50% before the edge and 50% after the edge, but the ringing remains, as well on the upgoing edges as on the downgoing edges. I have seen that in extenso on output signals of CD-players. There the ringing can be enormeous long, but fully symmetrical left and right of the edge or pulse. It looks as if your active filter suffers from overload to the positive side..... Could that be possible? I have seen that too on CD-players of "unknown" brands.

In the last days I did a lot on the monitor. Almost all circuits got their place on the chassis, or below. The FM demodulator above the chassis is running, you have seen the demodulated oscillogram. The powersupply, below the chassis is also running. Today I installed the transformer and the on/off circuitry. No external power supply is needed anymore, it's running now directly from the 230V AC mains. I should make a photo of the current set-up to show here and for documentation. I am missing still a crucial stud to mount the picture tube, so I can't show light off the screen.

[Steve Anderson]

Thanks Klaas. The reason I started on the sync arrangement where seven samples were required to say it was truly syncs was precisely due to the undershoot in the demodulated signal causing mis-triggering. The number seven is nothing special, it just comes out as 1.5ms of continuous sync before anything happens.

Why the ringing is generally on the -ve going edges and not the +ve I'm not sure. There is some on the +ve going edges, but much less so. I very much doubt it's overload of the op-amps. The signal is only around 4V p-p and the supplies are +/-12V, 24V total for the demodulator output filter. In some places just the +ve supply is used, even so I don't think this is the root cause.

It's a shame that there's no arrangement which is half-way between a Butterworth and a Bessel filter, well none that I know of. A Bessel filter to remove demodulation ripple would possibly require 10 or more poles. It may be possible to build a suitable FIR filter in software but it could suffer the same result, but perhaps I'll look into it.

I'm hoping that all this effort pays off when I start on the NBTV version of the software. I'm hoping that it's mostly going to be a matter of adjusting timings and reading/writing to/from the RAM backwards. (I can dream!)

Seems you're making good progress on your monitor...

Steve A.

[Klaas Robers]

Oh Steve, there are filter solutions between Butterworth and Bessel. I have seen transitional filters, that are bessel until Fcut x 2 and then Butterworth. But for this application the area between Fc and Fc x 2 is so important.

I have from 1973 the "Handbook of Filtersynthesis" written by Anatol I Zverev. The most impressive part is the collection of amplitude diagrams and step responses and impulse responses of many types of filter designs. I have used that always as a menu to select what I should. But the outcome is always that the ringing is seen as to heavy (Chebychev and even more Cauer), or the stopband rejection is too poor (Bessel and Equally Tuned) . The compromise in the middle is Butterworth. For video this is a suitable solution as long as you add an allpass to spread the ringing from only after the edge, to both sides. Some ringing gives an impression of sharpness.

The ringing is essential. This is the negative of the frequencies that you are missing due to the suppression of the frequencies above cut off. Here you see the wave form of the spectrum that was removed. Remove the ringing and you have the suppressed spectrum back.

That is the reason that I suspect the absence of ringing on your transitions to white. This indicates that new spectrum is inserted, or part of the pass band spectrum is removed as well. I would try to connect a pure square wave to the input of your video low pass filter and watch the output. See that the DC values of the output correspond to the black to white voltages of your video signal. Then you might see how the filter itself behaves.

If that is correct the source of the missing ringing must be in the circuits before the filter. It is not impossible that the Q of the 2500 Hz, and/or 1200 Hz tuned circuit in the detector is too high and swings up and down too slow. That might remove the higher video frequencies in a Bessel way, e.g. the frequencies from 500 Hz and higher and make the edge "slower". The round shoulders on the transients towards white look also suspicious. I remember myself that the slope of an FM-detector should cover the whole spectrum including the side bands. For SSTV this would be 300 - 3200 Hz (+900 Hz at both sides). The Robot detector does not fulfill this requirement, which still says nothing about the usefullness for demodulation of SSTV-video. This is the twilight zone between science and engineering.
.

[Klaas Robers]

One thing that might be 'educational' is to do a baseband to baseband screen capture and see how much the mod-demod process impairs the picture...

You were thinking of this, Steve? Or a comparision of the oscillograms?
.

[Steve Anderson]

I had thought of testing the demodulator output filter in isolation to the rest of the circuit, it's quite simple to do, I just haven't got around to it yet. At this stage there are other things that are more pressing, like setting correct black and white levels for a start. Although yes, I'm getting pictures, there's still quite a bit to do.

I don't expect the baseband-to-baseband test to be as good as your example but it will be interesting to see the effects of removing the mod-demod process and the associated filters on the picture quality. A point of interest, no practical outcome is likely to result.

Tomorrow I'll be continuing to work on the sync extraction process and improving the vertical detection software. If that goes smoothly (hoping, hoping) then it's time to deal with those black and white levels.

The method of design is an iterative process.

Steve A.

I've just re-read your posting above, and I quote..

"...For SSTV this would be 300 - 3200 Hz (+900 Hz at both sides). The Robot detector does not fulfill this requirement..."

I agree wholeheartedly. But the robot 70 arrangement is a proven design which as it turns out isn't that bad a starting point when you consider that it's a late 60s or early 70s commercial design. All I've done is tarted it up a bit. I did mention somewhere in the dark past that I intend to replace the Robot 70 designed demodulator with something better...which is one of the reasons it's built on a separate board.

Something along the lines of this looks promising...

http://www.agurk.dk/bjarke/projects/dsp ... 20sstv.htm

...but that's some way down the road...and the division is a big bug-bear for most micros, here's where straying into 16-bit micros might become essential. They have division instructions whereas the 8-bit ones don't.

Steve A.

[Klaas Robers]

Thank you Steve for the link to KR7A. But I fear he makes a thinking error. He starts with assuming that there is no signal below 1200 Hz and above 2300 Hz in the FM signal and he cuts off everything above 2300 Hz after I-Q mixing 1.75 kHz down. Then modulation at these higher frequencies (near white) becomes single side band. You throw away definition I fear. But FM is a difficult theory, I also don't understand it completely.

On the other hand, video signals obey completely different rules than sound. Non linearity in a detector slope can be hardly visible in a picture, it is only a change in gamma, while for sound it is a disaster. A constant group delay is immediately visible in a picture, while in sound you don't hear it at all.

If I don't know which shortcuts I may make, because it doesn't hurt in a special case, I like to do things according to the theory. That is the reason that in 1973 I have chosen for the counting detector, running at the doubled frequency of the incoming LF signal. I think that also an active low pass filter could have been following the counting detector, but then we had the nice ferroxcube potcores to make a filter of. Not a commecial solution, I agree. But as an amateur you generally make only one of a kind.

[Steve Anderson]

The KR7A demodulator idea was one of those, "I'll come back to it later." type of sites, I haven't studied it in any detail yet.

For the time being the updated Robot 70 demodulator will do now that I've got around the mis-triggering due to undershoots. To be honest I'm surprised it works as well as it does. The output filter is of my own design (with a little help from Filter-Pro), I should go back and see what Robot actually did in their design...out if interest only, I'm not going to expend the time and energy building it when I'll probably be replacing the whole demodulator section in the future anyway.

I must get back to tidying up the sync detection and the black/white levels.

Steve A.

[Steve Anderson]

After a few days of interruptions I've made progress with the black/white levels. Currently only eight bits of video data are being used, I need to add a few lines of code to bring this back to 10-bits.

I also need to start looking at the interpolation aspect of this. First-up will be simple line averaging where the new 'phantom' line is derived from the average values of the preceding line and the following one. Not a particularly good algorithm, but maybe better than the simple line-doubling done currently.

Test charts are bound to look worse in the vertical dimension, but real pictures may look subjectively better. Note the word 'may'.

Bi-linear and possibly bi-cubic to follow.

Below is the original Epsom pub photo and the black/white level corrected version. Also a shot of a cottage in Shere, Surrey. I still have to sort out the junk in line 1 (binary 0).

Steve A.

[Steve Anderson]

Though I'm still working on the SSTV part of this device my thoughts are turning to the NBTV-625 up-conversion part.

I have a small dilemma w.r.t. aspect ratio. I say small because many would think I'm be pedantic.

Correct me if I'm wrong but the NBTVA standard for aspect ration is 3:2 (V:H)...I think, but unusually this includes the sync duration rather than just the image shape. Referring to the nbtv.org site under "Standards and Recommendations" sync pulses are specified as being 0.1ms to 0.25ms. The the nominal 400Hz line-rate means the active video part of the signal can vary from 2.25ms to 2.4ms. It seems a small variation but it could well be visible on a circular shape...in fact the SSTV part still doesn't have the correct aspect ratio so maybe I should not bother with this too much, though I do intend to correct it in the future.

What I'm driving at is that in the NBTV case the visible aspect ratio could vary from 2.88:2 to 2.7:2...I'm hoping my sums are correct!

Perhaps I should just go with the sync pulse duration of 125us as I always use here and be done with it?

I did use the word pedantic...

Added a few hours later...

When I do get to the NBTV-625 up converter I'll start off a new thread in the NBTV section as it's relevant there. There is largely no difference in the hardware. In fact I imposed a self-inflicted hardware design freeze on this sometime late July. For NBTV you will not need the SSTV demodulator but will need a DC-restorer (simpler) if your source is AC-coupled (they usually are). Though it maybe possible to do DC-restoration in software...I haven't yet thought much about it.

Steve A.

[Harry Dalek]

Well its might be something that could bother some the important thing is what it does and show an image ....also how much extra circuitry or programming is needed to make it correct...? it would not bother me if you want votes its still impressive ,

[Steve Anderson]

Thanks Harry, I know you're working on your own current project, I did wonder where you had gone...but no problem.

I'm waiting for the sun to go down so I can take some more video of it in SSTV mode. The office/workshop windows face north and west unobstructed, in the middle of the year the midday sun is in the north as I'm only 13 degrees north of the equator. Though now the sun is slightly south of its zenith at midday...it's still damn bright in here!

I think I have cracked the flicker problem between the 25/50Hz frame/field rate of 625 TVs and your average digital camera. I'll know in probably less than an hour. If there is a deity then 12.5Hz NBTV should also be flicker-free...well, it will still have the 12.5Hz update rate, far better than yer average web-cam.

More soon...

Steve A.

Having entered the above I now hear the distant rumblings of thunder, if it's not one thing it's some other!

[Steve Anderson]

With the thunderstorm impending things got dark quick!

So 625 to AVI, no flicker.

First is a lane in Shere, Surrey, second is The Albion Pub in Epsom High St. Surrrey and lastly The Leg of Mutton and Cauliflower Pub, Ashtead, Surrey.

Steve A.

If anyone has a problem in viewing the video please let me know. Unusually Ubuntu Linux doesn't like it...a different camera to previous, but VLC on WinXP SP2 is fine...as is 'Doze Media Player...which version, no clue...