New Coleco RGB board?

[ChildOfCv]

12 hours ago, Falonn said:

- Ch2 (cyan? U2A output

Yeah, cyan IS a sad color.

12 hours ago, Falonn said:

Test conditions (in order):

- solid "white" (actually fuscia) screen

- solid "black" (actually red) screen

- Popeye mode select

 

  

This is a classic example of open-loop gain, where feedback is not pulled back into the negative pin.  Op-amps have enormous open-loop gain, so even a tiny difference in input voltage sends the output voltage into the rails.  In this case, it's almost always positive, except when the sync spike disturbs the Force.  Check the resistance between the "-" pin and the output of "U2A".  Put both probes on the IC pins.

[Falonn]

3 hours ago, ChildOfCv said:

Check the resistance between the "-" pin and the output of "U2A".

The meter is showing 740 ohms across the IC pins while in-circuit.  The feedback resistor (R11) tested in isolation afterward is exactly 1000 ohm, down to as much precision as my meter has.  (I intentionally grabbed a set of 1% tolerance resistors for this project and then hand-picked the closest one from the 20pcs of each value as I went along.)

[ChildOfCv]

5 hours ago, Falonn said:

The meter is showing 740 ohms across the IC pins while in-circuit.  The feedback resistor (R11) tested in isolation afterward is exactly 1000 ohm, down to as much precision as my meter has.  (I intentionally grabbed a set of 1% tolerance resistors for this project and then hand-picked the closest one from the 20pcs of each value as I went along.)

The IC is definitely not acting as expected then.  The datasheet shows its open-loop gain is about 32000 times amplification, so even a 0.001V difference would be multiplied to 32V output (which is clipped at the 5V rail).  If that 5V were really making it back to the negative input, then it should be pulling the voltage there closer to 5V, until it comes within microvolts of the positive input.  At that point, the output voltage would be stabilized at the positive voltage input (plus the intended feedback gain).

 

The circuit is basically this:

 

While U4B is closed, output should track the potentiometer input, such that output is 2.5V+1.5*(R-Y-2.5V).  When U4B opens, output should return to 2.5V.

 

There could be components contacting where they shouldn't, or components that just don't make contact.  What I'd do is attempt to isolate the circuit until that part of it works.  For instance, remove R14 and see if it becomes a pure voltage follower.  Maybe patch out the input signal and run it to a potentiometer like in my example above.  Or remove the output's link to the color summing circuits.  But R11 is vital to the operation of this circuit.

 

Edited March 27, 2020 by ChildOfCv

[Falonn]

Well, that's anti-climactic.  Last night I had swapped back to the first XR8054 SMT-to-DIP adapter board I'd made for no other reason than because I'd already added the little "U1 & U2" label to the first one.  Tonight, in preparation for my anticipated long debugging search (possibly rebuilding the whole board if necessary), I did a quick power-on to make sure things were in their last known state... and this was the result:

 

 

So, I suppose, lets chalk it up to a flaky breadboard and hopefully it won't crop up again.  *sigh*

 

Alright, first off: lots and lots of ringing!  Red takes too long to discharge.  Green areas begin with too much overshoot.  Some of this is probably the breadboard's fault, but I'm going to try the same capacitor trick from the TMS board to see if it can at least mitigate the green problem.

 

As for the good: the colors are (a lot) better than ColecoRGB 1.2 and it doesn't suffer from the awful cut-off pixel problem.

 

Comparing it against the CoolCV emulator screenshot (which I haven't actually validated as having accurate colors), blue is a little dark and green is super bright, almost neon.  The difference between the two colors of grass is hard to distinguish because they're both super saturated.

 

Question 1: Does anyone know if there is a nice, reference standard emulator out there that has gone out of their way for good color accuracy?

 

That would help with verification and tuning.  I've seen this AA topic, which seems to take the subject pretty seriously, but I'm not sure if that work ever ended up in something I can download.

 

Also last night, while trying to dig into op-amp theory (for the third or fourth time) and trying to figure out the operations for each amp, I got a different answer for U2A's output than your result above, ChildOfCV.  It was tough to track down anyone talking about negative feedback going to anything besides ground, but this answer looked promising.

 

With Vi=2.5V (Vref), Ri=2k, Rf=1k, Vp=R-Y, and using their formula, I got: 1.5*R-Y - 1.25V.  Although, with an unknown impedance on the R-Y line, I wasn't quite sure how to calculate (using superposition? or the op-amp summing network formula?) what to do with that 10k to Vref.  Still, I'm curious how you found that the 1.5x also applies to the negative feedback Vref term?

 

In many topics here (and several times in this one), I've seen people describe the TMS992xA as outputting something like "almost component but with weird scaling and biasing".  While digging around some more (last night was research-heavy), after re-reading "in conformance to standard industry practice [we divide by 1.14 and 2.03]" in the TMS doc, I was curious how that was standard.  In the Hackaday schematic, he even named the B-Y input line "U".  Looking up the YUV color space, it's apparently exactly the same as component (YPbPr, which uses B-Y and R-Y lines) except with those scaling terms.

 

Question 2: Does that mean a more accurate description would be "YUV with a DC bias" and that we should be calling these the U and V lines instead of B-Y and R-Y?  (Was that just nomenclature that wasn't in use when the chip was designed?  Do they avoid it because YUV seems to also imply half-bandwidth for the U and V channels?)

 

Once we're free to search for "YUV to RGB converter" instead of something much more TMS-specific, there are hundreds of these things all over the place.  One of the first results is this delightful project with a wonderful schematic.  The op-amp analysis is already done in the margins and it looks like much(!) more care was taken in picking the resistor values than any of these three other known boards.  (The schematic continues to go to the trouble of listing the parallel resistor combinations needed to get the exact values required.  It's really very nice.)

 

That reminded me of the wording in the TMS doc: "[Our] purpose is to produce a bright, sharp image [...] rather than to produce a very accurate color difference..."  And now I'm wondering how much better things could get if our purpose was actually color accuracy! ?

 

I haven't studied it in detail yet, but the idea in the (what I'm calling) ELM circuit above appears to be a combination of all the good parts we've seen so far: the same DC restoration idea from the TMS doc (and possibly color burst cleanup at the same time by using an LM1881?!) using a mux the same way as TMS Doc uses the 4066.  And then two amp stages like the Hackaday board, which should allow for much better color accuracy (when you actually take the time to pick good resistor values).

 

It does still need a -5V rail which is kind of a bummer, but in doing so, it saves all the complexity of the sample-and-hold for the bias voltage (and makes the analysis almost trivial).

 

I'm tempted to build the ELM circuit as-specified, like the others... but I wonder if just grafting those nice resistor values along with our other lessons learned as we near the final solution wouldn't be more expedient?  Now that it's been, what, two months that we've been dissecting these things, I'm starting to feel a tiny bit more confident about what to expect from this repeating pattern of "a little up-front cleanup followed by voltage math".

[ChildOfCv]

On 3/28/2020 at 11:55 PM, Falonn said:

Also last night, while trying to dig into op-amp theory (for the third or fourth time) and trying to figure out the operations for each amp, I got a different answer for U2A's output than your result above, ChildOfCV.  It was tough to track down anyone talking about negative feedback going to anything besides ground, but this answer looked promising.

Most op-amp theory expects to have a negative voltage on the V- supply and a positive voltage on the V+ supply, and 0V for the center, where gain applies with inputs above and below 0V as a simple multiplier.  We can still do that with this circuit by re-labeling the voltages.

 

Instead of calling it +5, Vref, and 0, call it +2.5, 0, and -2.5 since Vref is close to 2.5.  Once that is done, you can consider any input to the color difference channels to be related to Vref.  So for instance, the color burst period would read around -0.4V relative to Vref, so if it were to go through the gain circuit, you would see 1.5*-0.4V, or -0.6V output relative to Vref.

 

On 3/28/2020 at 11:55 PM, Falonn said:

With Vi=2.5V (Vref), Ri=2k, Rf=1k, Vp=R-Y, and using their formula, I got: 1.5*R-Y - 1.25V.  Although, with an unknown impedance on the R-Y line, I wasn't quite sure how to calculate (using superposition? or the op-amp summing network formula?) what to do with that 10k to Vref.  Still, I'm curious how you found that the 1.5x also applies to the negative feedback Vref term?

First, always note that for impedance calculation, all power supplies are effectively a short.  That is, if you were to "measure" the resistance of the power supply, you'd ideally get 0 ohms.  One example of circuit analysis with this principle is the common emitter amplifier.  Base impedance is calculated by putting both base resistors and the equivalent resistance through the base of the transistor, all in parallel and calculating that.  It doesn't matter that one of the resistors goes to VCC, because VCC is shorted to ground for purposes of circuit analysis.

 

Anyway, we have the 1K resistor on negative feedback to the "-" input, and a 2K resistor from there to Vref, which we now call 0V.  So the feedback is a voltage divider that supplies 2/3 of the output voltage back to the "-" input.  So in order to balance the "-" with the "+", it needs 3/2 the output voltage, or 1.5X.

 

The 10K resistor on the + side will slightly pull down the input signal, but its value was chosen to be a negligible influence, yet force "R-Y" and "B-Y" to neutral during the color burst.  The TMS inputs are supposed to have 390-ohm resistors on each line.  If you're connecting the Hackaday circuit directly to the TMS9928, you should also have 390 ohms pulling each of the 3 components to real ground (not Vref).  Well, if the TMS is still in the Coleco, though, then it already has those pull-downs before the analog switches anyway.  So maybe forget I said anything about that...

 

Now, the next question is, which op-amps are giving you the undesirable signals?  The first set that squelch out the color burst, or the second set that turn it into RGB?  If you have a TV with YPbPr input, you should be able to take the input directly after the first set of of op-amps for comparison, to see if it's the source of the signal issues.

 

On each first-stage amp, you can also place a potentiometer in place of the two feedback resistors, where the wiper goes to the "-" input.  Then you can adjust the relative amplification of each op amp.  For the second stage, it seems that trimmers would have to replace individual resistors in order to adjust for proper color balance, since many of the colors are derived from the sum of 2 or 3 of the input channels.

 

But all in all, that's a pretty picture, even if the red has to bleed in.  After looking at RF, I'd be quite happy with this.  And yeah, it could be left-over resistance from the likely bad connection you were seeing before that's making the red so slow.

 

On 3/28/2020 at 11:55 PM, Falonn said:

Question 2: Does that mean a more accurate description would be "YUV with a DC bias" and that we should be calling these the U and V lines instead of B-Y and R-Y?  (Was that just nomenclature that wasn't in use when the chip was designed?  Do they avoid it because YUV seems to also imply half-bandwidth for the U and V channels?)

Yeah YUV, YPbPr, Y with R-Y and B-Y, etc. are all pretty close to synonymous.  It describes a system where there's a luminance component and two color coords that give the appropriate dot a hue.

 

[Tursi]

I know you have some good experts here, but I found the EEVBlog primer video on OpAmps really informative...

 

https://www.eevblog.com/2014/04/06/eevblog-600-opamps-explained/

 

[ChildOfCv]

One more note:  The Hackaday project uses 47uF capacitors for coupling the final output.  I've heard it said that you need a lot more to avoid attenuating full bandwidth.  Funny enough, I once checked out a Sony VCR schematic and it used 470uF capacitors for coupling.  So it may be worth testing with 470's to see if that improves the picture even more.

[Osgeld]

I have always used 100 to 220 uF for video outputs, with the defaults typically being 100uF, course not with coleco and RGB and not generic op-amps but yes 47 in general seems quite low just for ntsc

Edited March 31, 2020 by Osgeld

[Falonn]

Thanks again, everyone.  As I seem to have jumped into the deep end of the pool in the hopes of learning to swim, all of you have been making patient, encouraging, and helpful gestures at me from the poolside while I thrash around and try to get my bearings!

 

@ChildOfCv I wouldn't have thought to use a different reference point.  That is a useful tool.  After carefully walking through your explanation (several times, slowly, hehe), it's all making more sense.

 

@Tursi I recall seeing that EEVBlog video a few years ago near the beginning of my self-teaching.  Almost all of it sailed over my head at the time.  Thank you for the reminder!  This time around I got a lot more out of it.  In particular, the idea of "virtual ground" was still in the hazy-concept stage in my brain.  Now it's a lot more concrete.

 

A few more results/notes:

 

1. I've been meaning to say for a few posts that for whatever reason, these screen captures seem to show a nicer, more forgiving picture than the TV itself.  In the Hackaday output, there was some distinct noise (almost "jailbars", especially on an all-black screen) and the top of the screen was brighter (with a white background cast) than the rest.  But none of that was showing up via the capture device.  Just now, I think I discovered the reason: viewing angle!  From my little workbench, I'm super off-axis from the TV (say, 150-degrees), so I'm apparently viewing with "enhanced artifacts" goggles on.  This was helpful, because I could verify an even greater difference when...

 

2. I tried bigger caps and it makes a huge difference!  470uF electrolytics are comically large (and a little scary) on the breadboard, but the white at the top of the screen and the repeating pattern in the solid colors (visible as jailbars on a black screen) are both completely gone.  Even from my severe viewing angle, things are very clean now.

 

3. Using the same trick from the TMS schematic of adding ~10pF capacitors in parallel with the negative feedback path on each output amp has cleaned up maybe half of the rising pixel overshoot, which is another subtle improvement.

 

4. I still haven't found a workaround for the falling pixel "smear"/slow-discharge.  It's nearly two pixels wide, which is kind of a deal-breaker (and weird, considering the XR8054 has ~3x the slew rate of the LM318, which doesn't exhibit this problem).  Although, I'm wondering if it isn't a moot point since it's most-likely opamp-specific and this model has been discontinued.  Or maybe it's still a weak breadboard connection someplace, like ChildOfCV suggested.

 

5. Switching back to the TMS board for one last experiment: I wanted to try the scheme from the (newly discovered) ELM schematic where DC restoration is done to B-Y and R-Y during the LM1881's burst output (instead of just during sync, like the TMS doc shows).  It's very close now, but there's still a tiny dip lower than the black level.  (And now the leading edge of pixels are very slightly softer, maybe 10% of the way toward the problem exhibited by ColecoRGB 1.2.  I suspect I didn't adapt the technique to the TMS schematic 100% correctly... there are a lot of moving parts in there and my attempt was a little hasty.)

 

So here is the current state of the project:

 

TMS Doc + LM1881 (and an inverter to make the Burst output active-high for the 4066B control lines)

Good: Very clean (soft'ish) output

Bad: Too much blue; requires oscilloscope tuning; requires +12V and -5V rails

 

 

ColecoRGB 1.2

Good: ... it outputs a picture

Bad: It's an unusably bad picture; requires oscilloscope tuning; full range of trimmers isn't enough

 

 

Hackaday + 10x larger output caps + small negative feedback caps on R, G, B output amps

Good: Clean, sharp output; no tuning required

Bad: Uses a discontinued part; some overshoot and color smear still remain; those nearly-indistinguishable greens make me worry about the color accuracy

 

 

Future work

- Try building the ELM circuit?  (I'm still really excited about those resistor values for color accuracy!)

- Find a replacement amp for Hackaday and hope that it also fixes the smear/bleed?

- Build the Hackaday circuit (with improvements) on protoboard (or even a proper PCB?) to determine how big an impact the breadboard is having?

- Any other suggestions?

 

What do you guys think?  I'm running low on experiments and we haven't found a perfect (or even almost-perfect) solution yet.

[MrPix]

Are you taking off the signals from the 9928 and disconnecting those pins from the coleco board? Or are they also feeding the coleco board still? Components on the coleco board would have an effect on driving your system if they're still connected.

 

 

[Falonn]

Since none of this will work with video generated by ColecoVision Expansion Module #1 (where RF will still be the only fallback) and for ease-of-installation, I fear that keeping the rest of the circuit undisturbed is a hard requirement.

 

Would it help if the very first thing we did was run each of the 992x inputs into a voltage follower?

[MrPix]

When Matthew releases the MKII he plans to open source the MKI - and that might be a more 'satisfactory' option - I would certainly spend the time to redesign the board using more higher level techniques (components both sides) to make it more general.

Until then (might be years) this work is important and I look forward to designing an open board for the winner.

[ChildOfCv]

You could try the MAX4392 (2 channel) or MAX4395 (4 channel) as a replacement.  Both are intended for video applications and support single supply of 5V.

[Falonn]

You caught me minutes before my next DigiKey order.  Nice timing!  I threw a few of each of those in there along with a grab bag of other things I've seen mentioned elsewhere and the couple of ICs for the ELM circuit.  I even included one of the (exorbitantly-priced) LMH1251 you can just make out in the picture of the RetroTINK COMP2RGB.  That's 2/3 (and the most interesting part) of the resistor network in a single IC, which might be interesting to experiment with.  If it doesn't reduce the cost, it would at least reduce the number of line items in the BOM pretty significantly. ?

[Falonn]

My Digi-Key order is supposed to arrive tomorrow.  In the meantime, I've found some interesting tidbits that might be helpful:

 

• The author of this (unrelated, but awesome) Colecovision-from-scratch project was struggling with how to get a -5V rail (for the controller ports) from just his 5V supply.  He pointed out the TPS60403 as a solution.  At $0.93, a tiny footprint, and Vout = -Vin (for Vin between 1.6V and 5.5V), that little charge pump would be a neat answer to keep the install as simple as possible with one fewer wire coming in from a nearby pad (or farther if you've done the 5V VRAM mod).  In terms of designing a DIY kit that someone can buy pre-assembled, the extra dollar is almost certainly worth having one fewer installation/troubleshooting step.  Now (should the ELM circuit end up superior), I'm less worried about it requiring a -5V rail.

 

• I noticed in the datasheet for the (exorbitantly-priced) LMH1251 that the lowest resolution it mentions supporting is 240p.  I hope I didn't just spend $11 on something that has no chance of working! I suppose we'll find out soon.  At least this was one of the rare occasions where I only ordered a single chip without any backups.

 

• The ELM project page keeps paying dividends.  They've got a series of schematics there from the most-complete version (with RGB, composite, and S-video out), the first simplification (with clean RGB only), and the second simplification (leaving the sync pulses on the RGB lines).  I had initially linked the latter because the other two included a CPLD which seemed a little out of scope for our purposes.  But, going back and reviewing the first simplification, that CPLD is doing something really cool:

 

The schematic has two of those muxes.  The bottom one (U3) might as well be a 4066B with an unused channel, that is, 3x SPST switches all controlled by the same line (or, 3PST).  It's used exactly the same as in the TMS Doc for DC restoration, except on all three channels instead of just R-Y and B-Y.  Much more interesting is the top one (U2), which is behaving like a 3PDT switch.  For the entire blanking interval (from the beginning of the front porch all the way to the end of the back porch), it just feeds analog ground into the op amps.  All pulses, blanking, syncing, and bursts are completely erased!  Once the blanking interval is over, it switches back to the just-zeroed-on-black signal lines and things proceed from there.  All op-amp math is done with the black level at 0V, which makes the rest very simple.  From an aesthetic standpoint, this is my favorite circuit idea by a mile and it intuitively feels like it would be the most compatible with any equipment out there.

 

Of course, to do any of that, you'd need to anticipate the blanking period.  The LM1881 takes a few ns to respond to the sync from the Y input and doesn't have anything at all resembling a "full blanking period" signal.

 

The trick is that the CPLD has been taught to understand NTSC blanking so it predicts them instead of reacting to them.

 

My plan is still to build the most-simplified version and see how it turns out.  If too-much-blue is still a problem, I suspect a little microcontroller (say, the 6-pin, grain of rice sized ATTiny9 at $0.33) using its timer peripheral and a couple interrupts could perform the same trick just as well as the CPLD.  Although, this is one of those DIY-kit hostile things: it would leave users in a position where they might have to find a way to program their own chip.  I emailed Digi-Key to see how much their programming service is for a one-off purchase (say, if someone wanted to build the final board from parts instead of getting it pre-assembled from someone like MrPix or MobiusStripTech), but I haven't heard back yet.

 

If it comes down to a microcontroller, ideally we could cobble something together that was adaptive so the same code works for NTSC and PAL.  The code listing for the CPLD on the ELM page has a "263 for NTSC, 313 for PAL" comment in the source, which is undesirable.  Placing a premium on install procedure simplicity, making a choice based on your region is another accident-prone step that it would be nice to avoid!

[MrPix]

The LMH1251 is available from main sources for $5 or less. Shop around.

Instead of an ATTiny (33c) consider a Greenpak (50c) - it has multiple internal timers, highly configurable discrete logic, and there's a couple of other functions it could do too. It also has some configurability. If you went with some Greenpak family models, I'd be willing to program as many as needed for free, and make them available to whoever at cost price.

[ChildOfCv]

On 4/8/2020 at 11:40 PM, Falonn said:

The schematic has two of those muxes.  The bottom one (U3) might as well be a 4066B with an unused channel, that is, 3x SPST switches all controlled by the same line (or, 3PST).  It's used exactly the same as in the TMS Doc for DC restoration, except on all three channels instead of just R-Y and B-Y.  Much more interesting is the top one (U2), which is behaving like a 3PDT switch.  For the entire blanking interval (from the beginning of the front porch all the way to the end of the back porch), it just feeds analog ground into the op amps.  All pulses, blanking, syncing, and bursts are completely erased!  Once the blanking interval is over, it switches back to the just-zeroed-on-black signal lines and things proceed from there.  All op-amp math is done with the black level at 0V, which makes the rest very simple.  From an aesthetic standpoint, this is my favorite circuit idea by a mile and it intuitively feels like it would be the most compatible with any equipment out there.

I guess this is why you mention the -5V supply chip then   "Analog ground" is what the Hackaday project sets at the TMS's neutral voltage during the sync pulse, which is around 2.5V.  Since this version here uses 0V for analog ground, it needs a -5V supply for the op amps and analog multiplexers (switches).  One note about adapting this circuit for the CV though:  Don't add the 75-ohm pull-downs.  Those are intended for a generic component input through RCA cable, which by standard has 75 ohms impedance.  The CV already has pull-down resistors for the TMS's preferred impedance, so those 75's would add a huge load that it didn't plan on.  But now that creates another issue:  Because we are not doing the 75-ohm impedance to 75-ohm pull-down, we also don't need to double the voltages on the first set of op-amps.  So the Y should just be a voltage follower.  You'll have to re-calculate the resistor dividers for the R-Y and B-Y though.

[Falonn]

Yeah, I knew to take the input resistors off (since the signal isn't coming in from an impedance-matched transmission line), but the 0.7Vp-p detail had escaped me, thanks!  I had hastily convinced myself that the doubling there was to get the output into the right range (despite that obviously leading to an incorrect order of operations in hindsight).

 

This is the modified schematic I put together last night with those input changes, replacing the CPLD with the uC, and updating the mux parts to the nearest model I could find that was still available.  (A fun coincidence: I only had to transpose the digits of the year for the "modified on" date.  18 years later to the day.)

 

Because the stages are laid out so simply without needing to shift voltages into the supply range, the new resistor values are easy:

 

For 1.4x gain, R8 should be 188ohm (or 270 || 620).

For 1.78x gain, R10 should be 364ohm (or 390 || 5k6).

 

I'll get the schematic updated with those.  (Although, I suppose those values are starting to get a little low.  I wonder if I should run the numbers to make sure we'll still be inside the LM6172's max output current...)

 

The Digi-Key stuff didn't actually arrive until this afternoon, but I realized a few days ago that these builds always take several consecutive evenings.  So there wasn't any reason I couldn't use stand-in ICs in the meantime and just build around those.  All of that is to say I've got a breadboard full of 555 timers (instead of LM6172's) that is a good 40%'ish of the way complete.

 

Regarding the -5V requirement, you guessed it.  Again, aesthetically I really like this one (more than the elbow grease it takes to shift things into range in the Hackaday circuit), so I've got high hopes that the results will turn out really nice in this 4th attempt.  Now that I know the negative rail can be solved for $0.93 (in single quantities, no less) that's all the better.

 

The most recent switch from the "level 2 simplification" to the "level 1" schematic (despite its requirement of a "smart" component that can anticipate blanking) is kind of a "stretch goal", but a completely clean RGB out with all pulses removed seems like a nice aspirational goal for maximum compatibility.  If somebody needs them (say, for Sync-on-Green), it's a lot easier to combine things again afterward than it is to remove them.

 

On 4/9/2020 at 11:05 AM, MrPix said:

The LMH1251 is available from main sources for $5 or less. Shop around.

With TI's own 1ku pricing showing $5.38, I would be skeptical of anything I found for "$5 or less".  Are you sure they're genuine?  The $11 I paid was individual pricing, which I usually expect to be high... just not that high.

[MrPix]

11 hours ago, Falonn said:

With TI's own 1ku pricing showing $5.38, I would be skeptical of anything I found for "$5 or less".  Are you sure they're genuine?  The $11 I paid was individual pricing, which I usually expect to be high... just not that high.

https://www.rocelec.com/part/TEXTISLMH1251MTX-NOPB

[Edit: nevermind, minimum QTY: 50 - even though they offer pricing for 1-24 and 25-49)...

Edited April 14, 2020 by MrPix

[Falonn]

Too exciting not to leave a quick note before sleep: the LM6172 creates a gorgeous picture.  No fiddling with capacitors.  No overshoot.  Full-width pixels.  No offset.  Just beautiful.

 

I've got a few more hours of programming to get the timing right on this microcontroller.  (The chip takes longer than the whole blanking interval to call the ISR!)   After that I'll have a screenshot that isn't corrupt, but all signs point to this being very promising.

 

Once that's done, I'll get to test my YUV vs. YPbPr theory by changing the resistor network.  Then we can run a poll or something about which set of screenshots looks more like the "real" colors from the RF output.

[MrPix]

Very exciting.

 

[ChildOfCv]

15 hours ago, Falonn said:

Too exciting not to leave a quick note before sleep: the LM6172 creates a gorgeous picture.  No fiddling with capacitors.  No overshoot.  Full-width pixels.  No offset.  Just beautiful.

 

I've got a few more hours of programming to get the timing right on this microcontroller.  (The chip takes longer than the whole blanking interval to call the ISR!)   After that I'll have a screenshot that isn't corrupt, but all signs point to this being very promising.

 

Once that's done, I'll get to test my YUV vs. YPbPr theory by changing the resistor network.  Then we can run a poll or something about which set of screenshots looks more like the "real" colors from the RF output.

 

What's the purpose of the MCU?  Couldn't a simple set of gates do what it needs to do?

[Falonn]

The burst can be eliminated with just pin 5 from the LM1881 to the DC restore mux (U3).  But U2 is doing something more interesting: removing all sync information from the RGB lines.  To do that, you have to switch it before the sync pulse, which means a timer and some counting.

 

Again, removing all sync is a little above-and-beyond, but I'd be willing to bet there is some wacky PVM out there that won't be compatible otherwise.  It's not like it's a proper "sync on green" that we're outputting otherwise.  Each line just has whatever strange combination of Y is left afterward, so it feels a little broken to me.  (This may be my inner perfectionist speaking, though.)

[ChildOfCv]

40 minutes ago, Falonn said:

(This may be my inner perfectionist speaking, though.)

Hmmm.  Yeah that may be more than could really be expected.

 

But one thing that may help out here would be to take advantage of propagation delay.  Instead of (or possibly in addition to) multiplexing ground into the component inputs, multiplex ground to the outputs of the second stages of op amps.  So the sync signal reaches the output multiplexer first and it has time to switch over to ground, before the Y signal has a chance to make it through the op amps.

Edited April 16, 2020 by ChildOfCv

[Falonn]

I like that idea a lot.

 

Hmm, in the meantime, I just tried switching over to the burst pin entirely to test my "the uC is just gilding the lily" statement above and I can't seem to get rid of the last bits of the blue.  Adjusting the 680k---not present in the schematic---on the Rset pin of the LM1881 up to 750k and then 910k was an incremental improvement each time, but the datasheet recommends not going much farther than that or the burst pin will start to exceed the standard timings.

 

While I'm here: this is just a small detail, but the LM6172 datasheet says that 14ohm is the typical output resistance.  Does that mean 60ohm would be a better value for the output resistors than 75?