shmups.system11.org

Nice result! It definitely looks better than what I tested back then. Other than differences in the mod board design, the quality difference may also be due to the SHVC-CPU-01 board revision you used, since it probably has way better analog signal isolation than the GPM-02 board I have. The image looks a little under-compensated since the blur is still slightly visible in some areas though. I'm sure the sharpness test screen in the 240p test suite is a less flattering as well.

I think it's possible to improve the image further without resorting to the digital outputs, but I just have a theory and no practical implementation of said theory at the moment.

TY for sharing that info! I'm going to order a couple of Borti's bypass boards and try those on the end of the chain.

God, I really hope this pans out. There are so many people that would appreciate something to make their SNES' picture look better, without having to resort to getting a 101. The RGB output on the original SNES is really bad.
We all know that, but it bears repeating.

Just finished. Not the prettiest but it works. Seems to be a slight but noticeable improvement

Hi everyone! After a couple of years since I saw this mod for the first time, I finally decided to give it a try, and the result surprised me greatly. Of course it is not perfect, as there are some noticeable ringing artifacts here and there, but in overall terms, I think that the image quality is closer to a 1-chip than to a stock 2-chip. So, in opinion, the mod worth it.

For anyone interested in the details, I installed the mod on a japanese NTSC SNS-CPU-RGB-02. This board has a slightly different analog video path, as can be seen in the following image:



Between the PPU2 and the video encoder, there is only one PNP common-collector stage, instead of 2. The RGB signals on the multiAV come from the video encoder, instead of coming directly from the transistor. For this reason, I had to add the missing common-collector stage, using three aditional transistors. I used the SMD version of the BC857, which seems to work well. I also used LL4148 (SMD version of 1N4148) for the diodes, and replaced 300 ohm, 200 ohm and 2 kohm resistors with 330 ohm, 220 ohm and 2k2 ohm ones, which are easier to find.

For the PCB, I made my own home-made design. Here is the board installed (forget about the green wires at the IC, I accidentaly swaped IN+ and IN- inputs on two of the three amplifiers when designing the board):



I lifted the base of transistors Q1, Q2 and Q3, took the original RGB signals from the PCB pad, and reinjected the modified signals at the transistors bases. I also lifted PPU2 pin 3, to remove the diagonal chroma sub-carrier interference (this is something almost mandatory in this board revision).

Here are some comparisons before and after the mod. The console is directly connected to a Framemeister and the image is upscaled to 1080p. Sorry for the quality, I don't have any capture device, so I had to photograph the TV with the phone:






I hope someone find this information useful.

Looks good! It's interesting to see different board layouts for that circuit.

I took a break from messing with this project for a month and just got back to it last week. My first adapter board didn't function properly. Graphical glitches and freezing everywhere. Not sure if I damaged the PPU2 chip or it was just bad soldering or something else, but I have more from a parts pile if needed.

Wow! This thread is awesome! Thanks to all who have worked on this or tried this!

Just one question. How does RGB with and without this mod compare to straight S-Video? I've been using S-Video on a consumer CRT and it looks pretty good to me. But when I look at RGB pics of non-modded 2 chips like mine it looks like a step down. Maybe it's just the close up effect.

Any chance of this improving any more? Or is this the final form?

I’ve been meaning to write up my thoughts on the Japanese mod. Now that I gained a little bit of experience from designing my own circuit, I think I understand enough about circuits to make some comments and suggestions for improvement. Still, I’m not an electrical engineer, so I could be wrong or misinformed on some details. It’s going to be technical, but hopefully someone here will understand it:

1. The first PNP buffer is driving a rather heavy load. This is mainly due to the 300Ω resistor terminated to VCC. For the AC part of the signal, the equivalent parallel resistance seen by the emitter is 279Ω, or 300Ω || (2kΩ + (2kΩ || 3MΩ)). This probably puts too much of a load on the unbuffered PPU2 signal, resulting in nonlinear distortion. Depending on the transistor grade (FQ, FR, FS), the base input resistance can be anywhere from 33.5kΩ to 156kΩ, which is quite low. For example, the THS family of video buffers have an input resistance of 800kΩ, and op amp input buffers are typically in the MΩ to GΩ range. Raising all the resistor values by one order of magnitude (3kΩ and 20kΩ) should reduce distortion and draw less current.

2. The output voltage swing of the LMH6683 is within 0.8V from the rails (VCC and GND) under a 2kΩ load. This means that the signal needs to be within 0.8V – 4.2V, assuming VCC is 5V. The circuit applies a load smaller than 400Ω at the op amp output due to the feedback and gain resistors, so that may shrink the linear region further. A single PNP buffer voltage offset is typically around 0.6V, so it isn’t enough to push the signal safely into the linear region of the op amp, resulting in more distortion at the lower end of the signal. The solution is to either add a little more DC offset or choose a different op amp that can swing well below 0.6V from the rails. Maybe it would have been better to apply the circuit after both PNP transistors? That way, the current gain is much higher, and the DC offset is about 1.2V. There are a bunch of ways to approach this.

I think replacing the op amp with a rail-to-rail, unity-gain-stable type would eliminate a lot of issues. The LMH6683 doesn’t appear to be unity gain stable, and that’s why the signal is attenuated by 6dB and amplified by 6dB. The voltage dividers seem to be the cause of unnecessary loads on the circuit. With a unity gain stable amp, the two 2kΩ resistors can be eliminated and the 300Ω one replaced with a much higher value like 10kΩ. The stock circuit already has a 10kΩ resistor between the first PNP emitter and VCC, so you will only need to tap the emitter pin to get your input. Furthermore, the 200Ω gain resistor can be removed and the 200Ω feedback resistor replaced with 100Ω to compensate. This will reduce the parts count and current load due to the elimination of low-valued resistors terminated to GND or VCC.

Anyway, those are my thoughts. I’m still not quite sure how the overshoot-limiting portion of the circuit works (it starts from the 47Ω output resistor, follows though the 2 diodes and finally the 1000pF series capacitor). My own approach is a little bit different, but I don’t know if it makes that much of a difference overall. Hope this helps anyone who wants to tweak the mod further.

Unmaker, did you use 75R resistors for R11,R21,R31 to adjust the brightness? May I ask what moved you to populate those spots at all, when the instructions say to leave blank for 2-chip SNES models?

Has this been distilled down to a purchasable ready to install board?

No. At least, not by me. The problem is that while the video filter mod board is an improvement in the opinion of most people, it introduces the overshoot problem discussed earlier. I'm going to try out Borti's bypass board as @unmaker did, and see how much i think that helps.

But then it's a two-board solution. I haven't even looked at the schematics to see where else the post-filter video signal goes, so i'm not sure there's going to be an elegant solution. And if there is, i'm not setup to sell it. I'd leave that part to someone else.

I think the overshoot problem will be difficult to solve using pure analog components. For now, the easiest thing to do is to hide most of it though optimal digital sampling. I guess that's not a sensible option on a 15 KHz CRT TV though. Maybe some new innovations (sans the digital mod) can improve the signal better than the way it's being done at the moment.

I'm not exactly sure how adding borti's bypass board helps with brightness. The Japanese RLPF mod outputs at the same levels as the input, with some attenuation and distortion due to the issues I described earlier. It looks like unmaker simply terminated the signal with 75 ohm resistors to drop it to 700 mV because the THS driver has a fixed 6 dB gain. It would actually have to be terminated at < 75 ohms if the picture was too bright for him. I'm not sure exactly what's going on because it should've been at the proper levels to begin with.

Internal to the shvc-cpu-01 model, Borti's bypass board doesn't require the terminators on the amp inputs. 75R terminations are specified for use on 1chip models, and 1.2K on SNES2 models. Which is why i was confused.

Still willing to give it a shot, it shouldn't hurt anything.

I took a look at the instructions for the mod and it looks like it requires the removal and replacement of some motherboard components and different solder points. You can repurpose the mod to simply terminate the signal and amplify it back up if you understand the whole circuit well enough.

Well, i have to admit i don't understand it. What are you saying in particular though - use the 75R resistors for R11, R21, R31? Don't use them?

I don't mean to be a bother but would anyone be able to help me find the parts I need to build this board? I'm not particularly savvy enough to know exactly what everything in the BOM is, and searching for some of it on Mouser yields either no results or way too many for me to figure out what part I have to order. If anyone could point me in the right direction and link me what's necessary I'd really appreciate it, or if anyone already has pre-built PCBs I'd be interested in buying one.

Soldering is no issue for me, I can follow directions easily enough, but that's mostly where the scope of my knowledge ends.

Good question, as I'm in the same boat, wanting to mod my 2-Chip SNES and not having a clue of where to get the components. I guess we need to look through this thread again since I do think there are clues as to what components are needed for this mod.

I have 3 unassembled sets (including PCBs, but not including the transistors) that I can bag into kits and send out.

This includes 1N4148 diodes in the correct package size, in lieu of the oddball diodes used in the original mod.

You will need to BYO transistors, and soldering skill. Honestly it's not too hard with a pair of tweezers and some patience.

I'm in Australia. I'll post within Australia for free (I have prepaid bags), and overseas for a cup of coffee (just to offset the price of postage a little).

I've modded my SFC in order to test a variant of Torapu's proposal, by use of buttersoft's pcb and inspired by yoshiyuki's comments. I tested the lmh6683, lmh 6644 and tlv3544, with the last one giving the best results in terms of sharpness and lack of overshoot. The two replacements I tested for the 6683 were fast quad op amps that have better specifications in some important parameters than the original part, they are pin compatible with the 8863 (with one channel going unused).

The datasheet for the 8863 specified an overshoot rating of 20% (yikes!), while the 3544 is ideally suited for video applications (rail to rail input and output, exceptionally flat gain and phase over a huge bandwidth). The 6644 is similar to the 6683, but slower and without the overshoot issue.

It's clear the 6683 was chosen as one of the fastest three channel op amp that is availablea, and it's quite cheap, too, but it's certainly not ideal. There is no reason to limit the part selection to three channels, as 4 channel parts are practically all pin compatible.

Details are as follows.

1. I purchased components and boards for a total of 5 units from mouser and jlcpcb. Including the replacement transistors and various capacitors in there range of 20 pf to 1 nf.

2. Fully assembled three boards with components as specified by torapuy and buttersoft, one for each of the op amps that I've mentioned above. I wired these with male dupont tips, using pin5 for 5v directly to the IC and the left side of R4 for ground.

3. Removed the Q3, Q5 and Q7 transistors from the board,then set them aside for a rainy day

4. Soldered a set of wires from the base and emitter of each transistor to a female dupont tip. Also soldered a pair of 5v and ground wires to PPU2's bypass caps.

5. Tested all three boards with different op amps extensively. Results area as follows.

The pixel sharpness is significantly improved with any of the options.

The pixel edges seem noticeable noisy (temporally) with the 8863, this is easy to see in a crt, and can be masked somewhat by using optimal sampling in the ossc and adjusting the phase... Results with this are suboptimal IMO. I think the output isn't quite linear, either, but the previous artifacts led me to discard this as an option so I didn't test too much after that. The artifacts led to false colors being shown in the edges of adjacent pixels, quite distracting I would say.

Both the 6644 and the 3544 are a better choice than the 6883, with no hint of overshoot. Both appear very much linear and give practically identical results through the ossc. It's clear that the 3544 is sharper when observed through a crt, though.

Image quality is 9/10, smearing is very nearly gone, you can still see a hint of the slow response in some black-to-grey transitions if you know what to look for, but it's miles apart from an unmodded 2 chip in terms of picture quality. I believe there is still something to be said regarding using faster diodes with a lower voltage threshold, such as shottky diodes org maybe germanium ones. I'll be trying that at some point to see if shallow transitions can be sped up, too. The current configuration requires at least 0.3v or more in a transition to be sped up (given the 2x gain and the 0.7v forward voltage of the diode), so shallow transitions are still slow. With a germanium diode this would be down to 0.1v or so, at which point it should be close to perfect.

Having found my preferred op amp, I went with a 47 pf capacitor in each of Q4, Q6, and Q8, this seems to be enough to filter the last hint of noise without affecting sharpness. I don't use the LPF ñor.the Reverse LPF options in my ossc anymore. Output is so sharp you would think I was playing a mega drive.

My SFC was already heavily modded, with a substantial amount of fast capacitors at the input and output of the regulator as well as at the pins of the S Enc chip and 10 uF smd ceramic capacitors in parallel with each of c90 through c99 as well as component video output. There is also a 5v bypass from the regulator output to S Enc 5v pin through a wire. With this there is no hint of bright/dark vertical line, with no jailbaits either and practically no noise in the ypbpr outputs.

Modding this SFC has been one expensive endeavor, but it certainly has been worth it to me. In order to recover some of the investment, I'll be selling the other 4 boards, fully assembled (with transistors and wires) for 20 USD each, shipped via registered airmail (w/tracking) and the optional 47 pf capacitors. Let me know if you're interested.

Credit to Torapu for the original concept and circuit design, buttersoft for laying out the circuit in a board, and yoshiyuki for some inspiring posts regarding the circuit analysis.

@yoshiyuki, it is possible to have different amount of compensation in the upwards transitions compared to the downwards ones by making the diode configuration non symmetrical. I'll test this soon enough.

TL;DR The sharpness of the output is amazing with this mod. Use the tlv3544 for best results. It's a more expensive amplifier, but the performance speaks for itself. I recommend it for your 2 chip snes, the output improves so.kuch it's impossible to go back.

What's a good way of uploading pics so that I can link them here? Imgur?