The issue I described as “blue tint” turns out to have been a fault in my test system. I built another revision of the PCB including pads for capacitors, and it would seem I only needed to place a cap on the clock line to resolve the remaining signal integrity issues.
Full VGA Clock (25.175MHz) without any pixel errors / artifacts
Full brightness control via the contrast keys
Less than 100µA draw on the 10.5V inverter supply while display is off.
All electrical load is on the inverter supply wires and not the fpc.
250mA fuse to reduce risk of blowing the internal 650mA fuse in any unexpected failure modes.
Using the Hitachi transceivers maintains same electrical interface to the CT65535 as the OEM display. This may be an overkill step but it is working well so I am sticking with it.
Further verify there are no pixel errors at the higher clock speed.
Revise the PCB to simplify assembly and clean it up.
Update the 3d model for the new bezel to accommodate the new PCB.
This is just a quick update. This is more a journal to myself and as always I most say I do not know what I doing generally, so consider this as entertainment more than information.
TLDR: I have had some success using the Hitachi level converters and terminating capacitors. I believe this issue is all about transmission line theory, which I am aware of but do not understand so this is trial and error. I am ordering another revision of the board this week and starting to try and finalize the 3D model for the bezel.
I do still notice a blue tint in certain mode tests, I am not certain this is a hardware issue yet I need to investigate it more once these other issues are sorted out.
I continue to chase the goal of being able to drive the replacement display at the full VGA clock speed of 25.175MHz / ~60hz refresh rate.
To summarize the issues I have been facing, in various versions of the PCB’s and hand wiring there was noise, jitter and other timing issues visible when certain patterns were displayed on the screen. The effect was not as bad when running at 15MHz but still there. It was terrible with sloppy handmade boards and breadboard testing.
I attempted to fix this with various forms of shielding, however it became clear the the issue was not external noise, and the reason the the metal back plate improved the situation was due to some kind of inductive coupling or similar effect on the ribbon cable.
I concluded the issue was related to an impedance mismatch causing reflections of the signal. I looked at what type of source termination and receiver end termination may have been in place.
I found that all signals including the clock appear to have a 0ohm resistor between the CT65535 output pin and the FPC connector, and they are each coupled to ground with a 50pf or so capacitor. On receiving side they went straight into a Hitatchi HD151015 9 bit Level Shifter / Transceiver.
I figured these values and topology were designed based on the length of the path from the CT65535 through to the CSTN at the original clock speed (~8mhz if i recall).
Of course this is no longer the case with the adapter board where two new connectors are introduced, the traces on the adapter PCB, the ribbon cable on the AT050TN23v1 and then the difference of the input gates/etc on the TFT.
I had various attempts at implementing signal termination at the converter PCB and on the TFT DCLK pin itself with varying success but nothing repeatable. What would work with test leads would not work once soldered on. What worked in one orientation would not work in another it was all random. I believe this may have at least in part been related to trying to terminate in the middle of the transmission line and not close enough to the end.
It had also been observed that lowering the voltage of VCC on the TFT could improve the issue slightly, this was not practical but got me thinking about level converters to address the flight difference the PC110 ~3.5V and the 3.3V of the DC-DC. I also knew that the immediate input on the OEM display was a Hitatchi level converter.
I purchased every possible level converter I could find on digikey, and generally the results were identical to it not being used at all. I then tried harvesting the Hitatchi level converters off an old CSTN and suddenly the signal was drastically improved, but still had some issues which I assumed was related to the hand wiring used for testing
I produced a PCB to properly mount the Hitachi chips and upon initial testing there was almost identical issues with the signal/display as without the Hitachi chips. This was starting to solidify a sense of defeat.
Due to missing (still missing) digikey delivery I was without the various capacitors to mount on the board, so there was still a small hope that the bypass capacitors on the level converters could help but not likely.
I then thought again if the previously attempted termination methods that failed would now have better luck given the long transmission line from the PC110 was now being terminated into the Hitatchi chip. Magically now, coupling the clock signal to ground with a 100nF capacitor at the input to the Hitatchi made the signal essentially perfect. Although I have not found issues yet, I expect there may be test images that result in some individual signal lines approaching clock speed and can benefit from the same filtering. Now I understand the issue is more about the speed of the transmission and not the frequency itself, but in practice it seems to be related to the frequency in this case.
I am going to order another revision of the PCB that will place 100nF capacitors as close to the Hitachi pins as possible. It is possible this approach may work with other actively available level converters but given the low volume of this project harvesting hitatchi chips from old displays and/or purchasing some limited qty of new old stock is good enough. It is also nice that the Hitachi chip has 9 channels allowing to accomplish the task with two chips, whereas most of the new parts are only 8 channel. The board is tight and my routing skills are limited.
Again in closing, I believe this all falls within “transmission line theory”, and with proper understanding it may be possible to simply place series resistors and/or capacitors on the converter board and with a correct value and solve the problem with less components i.e. no converters.
I am however taking the route of using the level converter chips to deal with the fact there is a difference between the Vcc from the 3.3V DC-DC and the “3.3V” on the PC110 motherboard. Putting the capacitors on the signal lines should create an RC filter, but I am not putting a resistor as I am assuming (from experimental testing) that the resistance of the trace/gate/etc is acceptable.
I am still trying to figure out what to do with the all of the signals but the clock signal in particular is of concern. It seems that the integrity of this signal may be the main cause of the distortion seen when running at 25.175MHz and even some at 15MHz. The distortion can be as simple as a single pixel shadow to the right on a black screen, or in more severe cases it appears as “jitter”. Previously we have attempted to compensate for this by lowering Vcc on the TFT display to try and have it handle the poor signal levels due to the assumed reflections. This signal integrity issue exists on all the signals, but the clock is the worst which is why I spent some time focusing on it. Maybe correction of it alone is not enough. Do I need to put level shifters/buffers onto the PCB?
I do not know what I am doing here, so if anybody knows better please help!
My measurement technique is almost certainly not correct and will be a contributing factor to the deformation/distortion of these measurements.
As we are looking for a solution that does not involve any modifications to the PC110 itself, there are only
CT65535 Drive strength paramter (XR6C bit2)
One complete frame is 525 lines. Each line is around 800 clock ticks. That means about 420k clock ticks per frame. So at 25.175MHz the refresh rate is around 60hz. At 15MHz it is closer to 35Hz. My goal is 25.175MHz still.
I already have the trace quite short, however I think I can do better to reduce the sharp edges and surround it better with grounds. I will do more reaching on this and try to look at other designs for examples.
Maybe this is too complicated and not needed, but maybe it is needed to use a voltage level translator much like what is found on the citizen board. I could place three of these 8 channel chips to handle all signal lines, so no longer a need to worry about the voltage difference between the PC110 “3.3V” and the TFT’s Vcc, as well the signal is regenerated.
Termination is a complex topic for me. I have no chance to calculate what is needed so it will be trial and error. I am contemplating how to place the appropriate pads to allow experimentation the potential termination methods (single parallel resistor, RC, Thevenin, Schottky diode) etc, but without further aggravating the issue of reflect with all those little unused pads.
I already experienced this issue, where I found a nice combination of capacitors (through trial and error) that made the signal perfect, however I was testing with long wires, when I then put the components onto the TFT it did not work because the situation was totally different without those long wires.
Depending on the termination choice taken, maybe it will be needed to increase the drive output of the CT65535 via XR6C bit 2.
Citizen Display Attached
Although waveform doesnot look ideal, you can see at least that the reflection is not so serious and it is close to 3V Vpp with minimum at 0.
The following is the clock waveform measured as it exits the hitachi level shifter on the CSTN control board.
This is what it looks like with no display attached
This is waveform with TFT attached.
PC110 DRIVING SIDE
The digital clock output from the CT65535 (Pin B8) connects to the LCD output connector Pin29 via a 0ohm jumper, and is coupled to ground with a 50pf capacitor. I would suspect the 0ohm jumper was a provision for driver side termination? there is no trace running under it so it was not for routing purposes.
On the Citizen display all signals are fed into level shifters.
The following is an updated version of the experimental vpatch. It reads from a config.txt for the values to update in memory, this is to make it easier to verify which values need to be updated without rebuilding the program every time.
the first line is a checksum, if you set it to 0 the program will not check and will update and reboot every time. If you set an expected checksum, it will only update/reboot if that matched, which is good for putting in autoexec.bat
the program has very limited error checking and very simple parsing of the config file so you have to be specific.
this program is the precursor to flashing the chip.
Progress is still being made on the exact modifications to be made to the video bios to get perfect picture in all video modes, so in parallel I am working to finialize the flashing patch so removal of the flash chip for programming is not required.
Currently to enable Vpp 12V for programming (pin11) I need to execute a sequence of read/writes to i/o addresses FC23h, F023h, C023h, 23h, 24h, 25h. The sequence was determined from disassembly of xpatch.exe
So the initial instructions on 23h do in fact seem to be an unlock sequence, and 24h,25h appear to be index and data registers. I had this feeling as xpatch was reading the values but doing nothing with them, so I started searching for the values with “unlock”.
The above patent document basically confirms the concept of four consecutive reads, to FC23,F023,C023,23 to unlock/enter a configuration type command. So this is some progress, it is good to know these do not have any special meaning. This may be common knowledge people who work on these systems often.
So this leaves to do more investigation on the 24/25 index/data registers, and actually to understand what is this even commanding, the VLSI chip or the CPU. I am in the process removing the BGA packages from a board and tracing the wires out.
I want to understand their exact purpose of each register and not “guess”, however the datasheet for the VL82C420FC5 “SCAMP” (Single Chip At Mid Performance) is not available so I continue to read similiar datasheets for individual VLSI components to understand the methodologies and interfaces.
The attached vpatch.c does not include the flashing routines, it requires more testing. This simply updates the bios in memory which is useful for testing the XR values to get it tuned in exactly where we want and verify functionality. Unfortunately I do not believe you can enter easy setup from a warm reboot.
When the program is run it checks to make sure the bios in memory is what it expects, and then changes the required values. A reboot (ctrl-alt-del) is required for the video bios to re-run its initialization routine. I do this as I am simulating if the flash had been updated. To make this a functional temporary patch it would be possible to set the XR values in the CT65535 registers on the spot as well, or potentially re-call the initialization routine.
The changes will stay in memory until the power is reset or a hard reboot / jump to FFFF:0000 occurs.
I am using my V3 PCB for testing. I have not modified the 3D printed bezel yet so it cannot fit flat as there is not quite enough space so I have just glued it standing up for testing and cut the window as needed.
I want to be sure the DC-DC I have selected is suitable before spending more time on the mechanical aspect of fitting it. I had one fail, the output capacitor failed to a short. This burned out the 630mA fuse on the PC110 inverter board.
On the next revision of the board I will be including a fuse below 630mA to try and prevent any faults in the TFT retrofit blowing the PC110’s inverter fuse as it requires the entire unit to be taken apart to properly replace it.
I currently have the onboard PIC10 set to control a fixed 5% backlight brightness while I investigate the DC-DC’s and what an acceptable max brightness is.
Trial Run Modified Bios
Here is a test run with a bios modified and programmed outside of the PC110 to demonstrate the end goal. The screen is not properly aligned as there are still a couple registers I need to properly update.
In a perfect world, I could buy a bunch of the keyboard/mouse adapter cables and just use regular devices. It is not easy to come by though, so I contemplated making a small dongle to allow pairing of bluetooth devices.
I searched quite a bit and have been unable to find this connector from any of the manufacturers, so perhaps it was truly a custom designed connector style. I was also unable to find it used in any other products, I tried to look at other old products by Ricoh that may have used it including the Thinkpad 220, but no luck.
As an alternative it would have been nice to do the straight PCB method that can be done with USB, however the contacts are recessed on the PC110 so contact would not be made to a flat PCB. What I contemplated was to use a spring contact on a PCB, along with an MCU. This is my proof of concept, mechanically it fits and holds nicely, and the contacts line up and make some contact so I will proceed to investigating what MCU to use and produce a PCB. As a fallback I will put a footprint for a standard ps/2 DIN connector so I could at least use a standard mouse. Not pretty but it will work.
I have done some work with Bluetooth/BLE a while back on another project, but this is new to me. What I need is a chip I can use to pair a bluetooth keyboard/mouse to, and then drive the ps/2 signals. Driving the ps/2 is the easy part, pairing to a mouse/keyboard is uncharted territory for me.