A New Method to Sharpen 3-Chip SNES Analog Video

The place for all discussion on gaming hardware
Post Reply
yoshiyukiblade
Posts: 63
Joined: Fri Sep 27, 2019 3:47 am

A New Method to Sharpen 3-Chip SNES Analog Video

Post by yoshiyukiblade »

As many of us know, all SNES hardware revisions before the "1CHIP" revision output blurry video, and there were a few attempts to address it. One attempt was by a hardware mod as discussed here (in the thread, you can see that I also tried my hand in making a mod, but I was never quite satisfied with the results). Another approach was by software, like the OSSC's Reverse LPF feature, but it still suffered from similar issues. Finally, a completely different approach was to use the "hidden" digital RGB signals tapped from the TST pins of S-PPU2. Opatus made tremendous progress on that front, and this may end up being the ultimate solution to getting perfect video out of the original hardware.

That said, I was still drawn to the analog output problem. It was clear that some amount of capacitance was causing the blur, but the amount of blur varied between different pixel transitions. For example:

Image

Here you can see a pixel transition from zero, to full scale, back to zero (0-31-0). The rise time of the pixel is so slow that it doesn't reach full scale in time (~186 ns), but it drops back to zero very quickly (~7 ns). This variable rise/fall time is what made the problem difficult to solve. Attempting to compensate the slowest pixel transitions would overcompensate faster ones, causing them to overshoot/ring.

To address the variable blur, we have to go one step further up the signal chain. The problem with our previous attempts was the fact that they were all post-filters. Once the signal is buffered, it outputs a uniformly low-impedance signal. Normally that's a good thing in most cases, but not here. A while ago, I theorized that the sensitive, unbuffered DAC output had a variable impedance which depended on the pixel transition. That, combined with the parasitic capacitance in the IC package, formed a low pass filter. If the pixel transition was high impedance, the LPF would be strong (low cutoff frequency, slow rise/fall time). If the transition was low impedance, would have a weak LPF (high cutoff frequency, fast rise/fall time). I searched for active compensation techniques for parasitic capacitances and came across the concept of negative impedance converters. If the parasitic capacitance in the IC was constant, it could be countered by injecting "negative capacitance" back into the signal. I built a prototype to test this theory, with promising results. The second prototype result looks decent enough to publish. Here's a screenshot:
Spoiler
Image
It doesn't turn the 3-chip console video quality into 1CHIP, but it might be good enough for some of us. I put some comparison screenshots of the Zelda 3 title screen in an album here. The first two screenshots are before/after with optimal sampling on the OSSC, and the other two are before/after with 8x sampling to see the fine signal details. All of them were integer scaled with nearest neighbor, with the correct pixel aspect ratio. There is still some blur left, but adding more capacitance starts to show overshoot on the OSSC with the 9 MHz LPF on. I think the slew rate of the video buffer I chose isn't quite as fast as I need it to be, so the overshoot recovery is a little slow.

Here are some scope plots showing the changes in the signal based on the amount of capacitance compensated. I used the 240p Test Suite Sharpness test as a reference for some of the slowest pixel transitions, mid-zero-mid (15-0-15) and mid-full-mid (15-31-15):

0 pF compensation:
Image

20 pF compensation:
Image

28 pF compensation:
Image

30 pF compensation (I settled with this):
Image

33 pF compensation:
Image

An overlay between 0 pF and 30 pF compensation:
Image

The compensation doesn't solve all the DAC issues though. There are still glitches that may be intrinsic to the DAC design. You can see some issues where 4-5 bits have to flip at once:
Spoiler
Image
Scope plots of the DAC glitches:
Image

Finally, here's a simplified schematic of my current implementation of the capacitance compensation circuit, with 75-ohm output (the video buffer LPF must be off). It was one of the simplest implementations I could think of:
Image

Note that it is not designed to work with the rest of the stock video circuit like the video encoder and AV multi-out ports. It is completely isolated and should output directly off the mod board. A fair bit of understanding of electric circuits will be needed to implement it to your needs. I think the SNR suffers a little due to the 2.9V reference instead of the original 5V.

Looking forward to seeing if anyone is interested in making a viable mod out of this. I may update the thread with a video playlist of comparisons I have recorded, but that will take a while. I'll continue researching for improvements, pushing the upper limit as far as I can.

Edit: I got a comparison video up. Watch in 4k to minimize color bleed from chroma subsampling: https://youtu.be/Hbk3wf0Ek8k
Edit: Created a playlist and added another video. More comparisons will trickle in over time: https://www.youtube.com/watch?v=Hbk3wf0 ... dR7-EDcP-I
Last edited by yoshiyukiblade on Sun Oct 16, 2022 5:24 pm, edited 3 times in total.
User avatar
Harrumph
Posts: 368
Joined: Tue Jan 19, 2016 10:06 pm
Location: Sweden

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by Harrumph »

Very cool! Thanks for your research and effort! To my eye, 28 pF seems to have indistinguishable rise/fall times compared to 30pF, but less overshoot. I’m a bit confused why you decided to go with 30 still.
yoshiyukiblade
Posts: 63
Joined: Fri Sep 27, 2019 3:47 am

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by yoshiyukiblade »

Harrumph wrote:Very cool! Thanks for your research and effort! To my eye, 28 pF seems to have indistinguishable rise/fall times compared to 30pF, but less overshoot. I’m a bit confused why you decided to go with 30 still.
It's hard to fully evaluate the image quality from the scope alone, but you can see where there is still a slow slope back up to the target level after overshoot recovery.

I believe there are even slower pixel transitions than the ones I used as a reference (even 30 pF isn't enough compensation). I wish I had the ability to make a test ROM that outputs pixel transitions between two arbitrary values to find the absolute slowest ones. Gonna work on getting some comparison videos up to show off the differences in motion.

Edit: Screenshot of the sharpness test with 30 pF compensation:
Spoiler
Image
You can see that there is just a bit of blur left on the black-to-gray transition.
User avatar
Harrumph
Posts: 368
Joined: Tue Jan 19, 2016 10:06 pm
Location: Sweden

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by Harrumph »

Well yeah it’s not perfect, and finding that optimal middle ground may be tedious, as you point out transitions differ from shade to shade. But still, judging from the zelda picture it’s like 95% there, and personally I’d take that over a 1-chip already.
yoshiyukiblade
Posts: 63
Joined: Fri Sep 27, 2019 3:47 am

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by yoshiyukiblade »

I'll keep testing with higher grade parts and more complicated implementations. This prototype was made as a proof of concept for a simple implementation of the blur compensation portion of the circuit. The THS7376 has 4 channels, low input capacitance, built-in 6 dB gain, DC offset, and good bandwidth/slew rate in bypass mode (no LPF), so it eliminates the need for a bunch of parts. However, the slew rate is still a limiting factor as its rise/fall times is limited to 8-9 ns.

The next test is to get the rise/fall times as fast as reasonably possible. Maybe I can get away with more blur compensation before the side effects start to become noticeable. Gonna see if I can get fall times down to 3.2-3.3 ns (over 100 MHz bandwidth). That's the fastest fall times I've seen coming out of the PPU's digital signals like the 3.58M pin, so I think it's a good arbitrary goal. Got another prototype in the works.
User avatar
NewSchoolBoxer
Posts: 369
Joined: Fri Jun 21, 2019 2:53 pm
Location: Atlanta, GA

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by NewSchoolBoxer »

Super cool post, thanks for sharing and being scientific! Side note for others: 3-Chip is another name for 2-Chip.

Is OSSC Reverse LPF an all-pass filter that tries to fix the phase change from LPF non-linear group delay? If so, good idea you may still want to use but, right, not helping with this.

(edited to be more positive and fix mix up of THS7376 with THS7374 :roll:)

So THS7376 has 375V/us slew rate and very impressive 300 MHz bandwidth at gain of 2 to make 7ns rise and fall times in theory. I zoomed in and added gridlines on page 27 of datasheet to measure it takes 13.75ns to rise from 0V to 1V. Once it gets going, it jumps from 1V to 2V in 6.25ns and that's in ideal test environment with no extra input capacitance to slow it down. Over- and undershoot not good. That was at +5V supply. At +3.3V supply on page 19, gains 1V in in ~3.7-4.7ns with no real overshoot.

Lower supply voltage giving less overshoot is expected but I'm surprised the slew rate doesn't suffer. Worth testing lower voltage with 3.3V regulator to make it better! Maybe 5V is faster on real RGB versus just ideal square wave input.

You should terminate the unused input on the opamp IC to avoid added noise or crosstalk power surging or, worse, oscillation. Is an oversight the other quad opamp mods should fix.
--------------------------------------------------------------------------------------------------------------------------------------------------------------

Curved rise is classic plot of capacitor charging, I can agree with your analysis of parasitic capacitance being the issue and generating a LPF. Excellent point about avoiding overshoot. I hadn't heard of the negative impedance injection strat and I only know of feedback around the (+) input for comparators. Pretty cool. I see other method of putting capacitor in parallel with gain resistor on the (-) feedback but alas not possible here and is only for IC's own parasitic capacitance.

Can move up out of $1 tier video amps. At $7.72 luxury tier we have the triple OPA3692 with ludicrous slew rate of 2000/us, bandwidth of 225 MHz at gain of 2 and GBW product of 200+ MHz. Actually lists rise and fall times. At +5 single supply, does 0.5V in 2ns and 2V in 2.3ns and I can eyeball the square wave pics and see a close enough match. The rise and fall times drop to under 2ns at +/- 5V, which is typical of opamps working a little better if you feed them the negative voltage even if they don't use it. No LPF but you know OSSC has one.

Risk with something that fast is instability that datasheet mentions can happen with significant capacitive load but have series resistor on output already to help with that. Actually, why do you use 2x 150 ohm resistors instead of 75?
--------------------------------------------------------------------------------------------------------------------------------------------------------------

If you want to push it to the limit, two fringe ideas:

-Cascade 2 opamps of same kind together to increase the effective bandwidth with extremely minor increase in noise. Gain of 100? Split into two opamps, each with gain of 10. Pretty ghetto to do with gain of 2 lol but let's see, effective opamp bandwidth you get now is GBW/2 = GBW*0.5. With two sqrt(2) ~1.4 gain stages, you get GBW*0.64 for ideal extra 14%. Better results at higher gain. Is old trick with quite a few online resources.
Obviously can't use the fixed gain video amps then. Can alternatively go [G+1.6] -> [G=+1.25] = gain of 2 or any ratio that multiplies to 2. Higher gain stage first reduces noise.

-Pricy 1 GHz bandwidth opamps exist but you'd have to buy three to add 3x2 gain resistors. Does give option of capacitor on (-) feedback to fight its own parasitic input.

Note: Circuit you want for custom gain is non-inverting to set gain by resistor ratio of [1 + Rf/Rg] so Rf=Rg for gain of 2. Datasheets will usually tell you optimal resistor values to use. Tend to be around 250-500 ohm.
yoshiyukiblade
Posts: 63
Joined: Fri Sep 27, 2019 3:47 am

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by yoshiyukiblade »

No idea how the OSSC's RLPF works. It was the first filter I used to sharpen the video of my SNES and was what sparked my interest in this problem (I didn't even know how to solder at the time, much less create circuits and design PCBs).

All 4 op amps are used in the THS7376, the 4th SD input was for CSYNC. I currently have it wired up for RGsB because I finally got fed up with SCART (poor/inconsistent signal integrity from the flimsy tab pins). Currently using the RCA jacks on the OSSC.

I took a look at the OPA3692 datasheet and saw that it has a very limited voltage input and output range. For a 5V single supply, the input voltage range is 1.6 - 3.4 V and and the output range 1.3 - 3.7 V. That isn't really viable for this kind of application. The output voltage swing of the SNES, using 5V at the AVCC pin, is 0 - ~2.45V. Negative supplies will help with the lower range, but not the upper range.

I use the 150 ohm resistor divider because of the output range I chose. See, the THS7376 has built-in 6 dB gain. With the 105 mV DC shift, this pushes the peak voltage beyond 5V (2 * (2.45V + 0.105V) = 5.11V), which is well beyond the op amp's upper output voltage swing. To counter this, I changed the AVCC input voltage to 2.9 V, resulting in a video output swing of ~1.45 Vpp. This lowers the SNR a bit, but allows 6 dB gain and DC bias without clipping. The result is simultaneously fed back into the input pin via capacitor for blur compensation, and driven out to cables. The voltage swing is ~2.9 Vpp from the 6 dB gain, so to get it back down to standard video levels, it has to be attenuated by 6 dB with a 75 ohm source termination resistance. The 2 150 ohm resistors accomplish that. With a 75 ohm load termination resistance, the final voltage swing is about 720 mVpp. It's slightly higher than the standard 714 mVpp, but still within the ballpark.

My op amp of choice is the ADA4855-3 for its good video specs. It was the op amp I used for the first prototype of this nature, but I ignored clipping at the time for simplicity. The slew rate is lower than the OPA3692, but its large input and output range is useful here. My next prototype will add a small negative voltage source to allow the voltage swing to hit GND and a little bit below. This op amp also has a very low input capacitance of 0.5 pF, allowing the total capacitance from the board to be less than 2 pF when you account for trace capacitance. With a slew rate of 870 V/us, or .87 V/ns, it can't get the rise time I want if I leave AVCC at 5V for maximum SNR. But if I leave AVCC at 2.9V, it can achieve rise times faster than 3.2 ns easily.
User avatar
NewSchoolBoxer
Posts: 369
Joined: Fri Jun 21, 2019 2:53 pm
Location: Atlanta, GA

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by NewSchoolBoxer »

OSSC sitting right on GitHub but no one writing up how it works. Looking at the BOM, I'm pretty sure the RLPF is done inside the FPGA. Nice to see what ICs a respected project uses when I've never shopped around for a Schmitt trigger buffer. Risk in copying a shopping list without due diligence. The designer living in PAL world, turns out the TVP7002 ADC video encoder doesn't compensate for jittery CSYNC from NTSC NES and SNES. Thus most of the demand for sync jitter fixing in those console mods. Only supporting YPbPr + RGB surprised me but I understand keeping design complexity and costs down.

What the encoder does have that I never though of is a LPF just for the sync signal:
SOG LPF SEL [1:0]: SOG low-pass filter selection. The SOG low-pass filter can be used to attenuate glitches present on
the SOG input. Excessive filtering can lead to sync detection issues and increased sample clock jitter.
00 = 2.5-MHz low-pass filter
01 = 10-MHz low-pass filter
10 = 33-MHz low-pass filter
11 = Low-pass filter bypass (default)

CLP LPF SEL [1:0]: Coarse clamp low-pass filter selection. This filter effects the operation of all enabled coarse clamps which is generally
the SOG coarse clamp only.
00 = 4.8-MHz low-pass filter (default). Suitable for HDTV and graphics formats.
01 = 0.5-MHz low-pass filter. Suitable for SDTV formats.
10 = 1.7-MHz low-pass filter
11 = Reserved.
Thanks for calling out the OPA3692 needs +5/-5V for good voltage input range. I should have noticed it's gimped on +5V single supply. Other super high speed opamp I'm looking at is LT6557 with impressive listed rise/fall time of 875ps and same 0.5pF input capacitance. But I see you're good with the 2.9V supply strat. Higher SNR with higher VCC, makes sense. I like your technical thoroughness. You know more about DC bias and video manipulation than I do. That's rough seeing input capacitance of THS7374 is 2pF on the datasheet.

I think ADA4855-3 or other non-video amp is good idea when you can fine tune the gain across each channel and then have 75 ohm termination in front of the inputs without forming a resistor network to keep Req.

Sending the sync through the amp too, that certainly makes sense to match the group delay and SNES is positive polarity. I'd think route it around in first design for not adding noise or consuming extra power but you know how to test if that is negligible or not.

I was wondering about negative voltage too. The ISL59837 video amp looks like fun with a small charge pump built in that can output -1V. I'm assuming basic non-regulated charge pump is sufficient here but switched capacitor converters are cheaper than I expected.

Sorry I didn't really help you at all. Exchanging ideas is fun for me at least.
User avatar
maxtherabbit
Posts: 1763
Joined: Mon Mar 05, 2018 4:03 pm

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by maxtherabbit »

NewSchoolBoxer wrote:OSSC sitting right on GitHub but no one writing up how it works. Looking at the BOM, I'm pretty sure the RLPF is done inside the FPGA. Nice to see what ICs a respected project uses when I've never shopped around for a Schmitt trigger buffer.
SN74LVC2G17 is a great part. I've used it in a few designs myself
yoshiyukiblade
Posts: 63
Joined: Fri Sep 27, 2019 3:47 am

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by yoshiyukiblade »

I am going to go with the full 5V AVCC to keep the SNR as high as reasonably possible. The 6 dB gain isn't a strict requirement; I can go with a lower gain (around 3 dB) and raise the compensation capacitor value. I never tried low gain, high capacitance before, so I don't know how it will look. It'll be an interesting experiment. The on-die noise is probably the most dominant noise source, so it doesn't make sense to make it worse if I don't have to. The output resistor network got a little more complicated, but it was still an easy calculation.

As for negative voltages, I searched for charge pumps a couple weeks ago and found the LM7705. This would have been the perfect IC because it was designed specifically for this kind of situation, a true zero amplifier. Unfortunately, it is out of stock basically everywhere and has lead times going into 2023. Gonna have to go with a more expensive and complicated adjustable charge pump in the meantime.
VajSkids Consoles
Posts: 123
Joined: Tue Jul 07, 2020 10:22 am

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by VajSkids Consoles »

try Ali express. you °might• get lucky.
2 x LM7705 for 12AUD

I went through this chip shortage annoyance with a common 3.3v LDO by micrel. Original back in stock date was end June (now) and then got pushed to July next year. That's on mouser, element14, Digi-Key etc.

ended up going on eBay and and getting a stack of them for a lot cheaper. Random places globally that had old stock on hand. They're all genuine too.
VajSkids Consoles
Posts: 123
Joined: Tue Jul 07, 2020 10:22 am

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by VajSkids Consoles »

$2.xx eBay and $2.xx postage but that comes anywhere from 2 weeks to 3 months to arrive
yoshiyukiblade
Posts: 63
Joined: Fri Sep 27, 2019 3:47 am

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by yoshiyukiblade »

Took a chance with eBay. Not the cheap ones listed though. Hope I get genuine parts. Gonna sit on the next prototype for a while and review it again before I send it off to the fab.

In the meantime, I'm going to work on the last "weak point" in the signal chain, which is the OSSC itself. After I got tired of SCART, I decided to practice my PCB layout skills by reworking the v1.6 board (not the latest v1.7 board because I have v1.6 here for parts). Swapped the RCA jacks with expensive 75 ohm BNC connectors, changed it to 4 layers, and changed all the 0603 resistors and capacitors to 0402. This lets me put bypass caps on the same side as the ICs for better power delivery, and add bypass caps to all the power input pins that are missing them. I also added more bulk capacitors with 47 uF tantalum polymers on all the the power sources, which may also help with low frequency power delivery. This should improve overall signal integrity and EMI by a lot, and keep the analog sections more isolated from the digital sections.

The stack up is signal+power/GND/GND/signal+power. Top layer also has a large ground fill to help with heat dissipation. All traces and planes are 8 mils separated from an unbroken ground plane, and pretty much all signal/power vias changing layers have a ground via as near as possible (~23 mils center-to-center or ~13 mils edge-to-edge). I was working on this board in May and June, and it just came in yesterday. Still waiting on some more parts next week. Hope it works because otherwise, I'll be out an OSSC for a while. Wish me luck.

I haven't described my entire test setup, but the S-PPU2 was desoldered from the motherboard and put on the mod board for maximum signal isolation. 75 ohm coax cables run straight off the board (routed through the hole of the RF module that I removed). Everything on board is powered by a dedicated 5V regulator.
VajSkids Consoles
Posts: 123
Joined: Tue Jul 07, 2020 10:22 am

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by VajSkids Consoles »

heaps of people will be looking forward to the results :)
yoshiyukiblade
Posts: 63
Joined: Fri Sep 27, 2019 3:47 am

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by yoshiyukiblade »

It took a while to assemble, but I finally got everything put together. Thankfully, it worked perfectly on the first power on.
Image
You may notice the bodge work on the input AC coupling capacitors to improve low frequency performance. I didn't fully account for it before putting the order in, but oh well. I'm using polarized 47 uF tantalum polymer capacitors in parallel with the ceramic caps, which is not ideal due to the risk of reverse polarity, but it's fine while powered on. I'll look for a better part later.

The results are looking great so far. I think there is a subtle quality improvement with less noise on the capture.

Here are some screenshots:
Spoiler
Image
Image
Image
Open up the screenshots in paint and use the paintbucket tool on the large flat colors. Some colors have very little noise. Some let you fill in chunks at the time. I think that's a good video quality test ;)

A nice benefit of a 4 layer board with 2 ground planes is improved thermal performance. On the original board, the TVP7002 had a heatsink. When I desoldered it, there wasn't even any solder on the exposed thermal pad. I think all the thermal vias wicked up the solder paste during reflow, so there was nothing else to aid with heat dissipation.

IMO, the project was a success. It wasn't cheap, but now I have a better tool for research and more confidence in high speed PCB design. Now back to working on the SNES mod.
yoshiyukiblade
Posts: 63
Joined: Fri Sep 27, 2019 3:47 am

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by yoshiyukiblade »

Time for a little update. A month flew by like nothing! Regarding my latest experiment that was meant to test the following:
yoshiyukiblade wrote:The 6 dB gain isn't a strict requirement; I can go with a lower gain (around 3 dB) and raise the compensation capacitor value. I never tried low gain, high capacitance before, so I don't know how it will look. It'll be an interesting experiment.
This gave poor results. I guess the feedback loop was unstable and it oscillated a ton. I had to use a much lower capacitance value than what was needed for good blur compensation. Funny enough, I couldn't see the oscillations with the OSSC's 9 MHz LPF on because they were at a high frequency (around 60 MHz). Attempting to add higher capacitance resulted in an output so unstable that the RGB signals started interfering with each other. The colors were all over the place.

I reworked the resistor network for 6 dB gain and had to wait for more parts to come in. So far, I tested the following capacitance values:

33 pF - Way overcompensated. Oscillations occurred in the LSB, MSB and a few other levels. On-chip noise was amplified to the point of being visible in some areas.
30 pF - Better, but still overcompensated. On-chip noise was mostly eliminated, but it showed instability on the scope. Without the additional input capacitance that the THS7376 has, 30 pF is a bit high on the ADA4855-3.
28 pF - Very clean, but still looks a little overcompensated. The LSB and MSB noise still shows noise spikes. It is still more compensation than 33 pF was for the THS7376.

One interesting observation is that there was still a tiny bit of blur visible on all the capacitance values tested above. It isn't viable to go as high as possible to reduce all the blur at the expense of overshoot. The best we can do is to limit overshoot as much as possible without increasing the existing blur. I may test lower capacitance values at some point, but I am leaving things alone for now.

Here are some screenshots with 28 pF compensation. They are the same as the ones above (though I can't get the same positioning for the Super Metroid one):
Spoiler
Image
Image
Image
Comparing the first 2 screenshots to the previous ones, the differences seem pretty negligible.

I'll be taking a break from this project for a bit, but I think I've probed the upper limits of what this compensation circuit can achieve.
User avatar
bobrocks95
Posts: 3461
Joined: Mon Apr 30, 2012 2:27 am
Location: Kentucky

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by bobrocks95 »

So are the screenshots in your previous July 9th post also a 3-chip with your circuit applied? Differences are negligible as in the 2 different approaches for the circuit came out about the same? I definitely don't see much of a difference between those 2 sets of screenshots if they're the 2 different circuit designs.
PS1 Disc-Based Game ID BIOS patch for MemCard Pro and SD2PSX automatic VMC switching.
yoshiyukiblade
Posts: 63
Joined: Fri Sep 27, 2019 3:47 am

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by yoshiyukiblade »

bobrocks95 wrote:So are the screenshots in your previous July 9th post also a 3-chip with your circuit applied? Differences are negligible as in the 2 different approaches for the circuit came out about the same? I definitely don't see much of a difference between those 2 sets of screenshots if they're the 2 different circuit designs.
They are the same implementation of two different circuits with different op amps used. The latter implementation is more complicated but uses a higher grade op amp. It does show faster overshoot recovery than the older circuit, but that is by looking at it through an oscilloscope. It doesn't seem to translate to a noticeably better output, at least on the OSSC.
User avatar
Harrumph
Posts: 368
Joined: Tue Jan 19, 2016 10:06 pm
Location: Sweden

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by Harrumph »

Thanks for your continued effort! Looking forward to when this can be had as a complete mod chip, would really love it in my 3-chip!
Ryeno
Posts: 89
Joined: Tue Nov 05, 2019 6:50 am

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by Ryeno »

yoshiyukiblade wrote:Time for a little update. A month flew by like nothing! Regarding my latest experiment that was meant to test the following:
yoshiyukiblade wrote:The 6 dB gain isn't a strict requirement; I can go with a lower gain (around 3 dB) and raise the compensation capacitor value. I never tried low gain, high capacitance before, so I don't know how it will look. It'll be an interesting experiment.
This gave poor results. I guess the feedback loop was unstable and it oscillated a ton. I had to use a much lower capacitance value than what was needed for good blur compensation. Funny enough, I couldn't see the oscillations with the OSSC's 9 MHz LPF on because they were at a high frequency (around 60 MHz). Attempting to add higher capacitance resulted in an output so unstable that the RGB signals started interfering with each other. The colors were all over the place.

I reworked the resistor network for 6 dB gain and had to wait for more parts to come in. So far, I tested the following capacitance values:

33 pF - Way overcompensated. Oscillations occurred in the LSB, MSB and a few other levels. On-chip noise was amplified to the point of being visible in some areas.
30 pF - Better, but still overcompensated. On-chip noise was mostly eliminated, but it showed instability on the scope. Without the additional input capacitance that the THS7376 has, 30 pF is a bit high on the ADA4855-3.
28 pF - Very clean, but still looks a little overcompensated. The LSB and MSB noise still shows noise spikes. It is still more compensation than 33 pF was for the THS7376.

One interesting observation is that there was still a tiny bit of blur visible on all the capacitance values tested above. It isn't viable to go as high as possible to reduce all the blur at the expense of overshoot. The best we can do is to limit overshoot as much as possible without increasing the existing blur. I may test lower capacitance values at some point, but I am leaving things alone for now.

Here are some screenshots with 28 pF compensation. They are the same as the ones above (though I can't get the same positioning for the Super Metroid one):
Spoiler
Image
Image
Image
Comparing the first 2 screenshots to the previous ones, the differences seem pretty negligible.

I'll be taking a break from this project for a bit, but I think I've probed the upper limits of what this compensation circuit can achieve.
Are you using the ADA4855 rail to rail or single rail? Do you have a link to where we pull video signals and power?
yoshiyukiblade
Posts: 63
Joined: Fri Sep 27, 2019 3:47 am

Re: A New Method to Sharpen 3-Chip SNES Analog Video

Post by yoshiyukiblade »

Ryeno wrote:Are you using the ADA4855 rail to rail or single rail? Do you have a link to where we pull video signals and power?
Those are two different properties. Rail-to-rail just means that the outputs (and sometimes the inputs) can swing very close to the voltage rails, but usually within 10s to 100s of mV. The rails can be VCC and GND for a single supply configuration, VCC to -VCC for dual supply configuration, or some arbitrary combination as long as it's within the total voltage swing allowed.

The ADA4855-3 is a single supply triple op amp with an absolute max voltage rating of 6V. I added a small -0.23V supply instead of GND so that the output signal can swing to GND and a little bit below. The total voltage is 5.23V, which should still be fine. However, it's kinda moot at the moment because I wasn't expecting to use 6 dB gain again, so there is probably some distortion near the rails anyway.

For my prototype, power comes from a dedicated onboard 5V regulator, which also powers the PPU2 chip. The stock power source should also work, but there won't be noise isolation. Video signals are from pins 95, 96, and 97 on PPU2.
Post Reply