With a successful integration test of our salvaged vacuum fluorescent display (VFD) we proceed to a few other tasks. On the hardware side, we’ll need to simplify our power supply situation. Our test had three separate AC plugs, we only really need one. We’ll also need to transfer what’s on the breadboard to a prototype circuit board.

On the software side, things are less clear because we’re still coming up with ideas on what to do with a salvaged VFD. Our very simple PIC driver code merely controls a pattern of bits, each bit representing a segment to be illuminated. This gives us a lot of flexibility but so far these bit patterns have been created using pencil and paper and quite time consuming. Once we have a project in mind we can write code specific to a theme, but until then it’s wide-open bits.

Well, at least we can streamline the pencil and paper system somewhat. The logic is not difficult, just time consume, and should be easy to automate. I started by creating a HTML form and laying out checkbox controls. I got as far as designing a JavaScript structure for checkbox click events when I realized everything is on a grid and looks a lot like a spreadsheet.

That realization led me to put my basic HTML and JavaScript on hold and switch to investigate what I can do in Google Sheets. The key that made it all possible is the BINTOHEX() function that reads a number, interpret it as binary, and translate to hexadecimal. This is the core functionality for my pattern generation and I was able to build everything else around it.

Here’s the initial build, with a fully populated grid of checkboxes and a test pattern for me to verify things were working.

Once verified, I deleted the checkboxes that didn’t have a corresponding segment. This array now properly corresponds to our specific VFD.

Now creating a bit pattern is as simple as checking on the segments I wish to have illuminated. Here’s the pattern for “12:00”, the canonical Hello World for a time-based VFD display.

I thought it would be fun to make this available for anyone to play with, but that requires keeping the sheet and its formulas fixed while allowing people to check and uncheck boxes. Unfortunately “view only” sharing does not allow checkbox manipulation, and “editable” sharing also allows modifying the sheet formula. Until I figure out how to do what I want, it was shared as “View Only” to the general public.

 

The time has finally come to put our salvaged vacuum fluorescent display (VFD) together with the prototype driver printed circuit board (PCB) we built for it. First we gave our VFD a little layer of protection. The individual pins of the VFD are small and seemed rather fragile. And as they went into a vacuum sealed chamber we had no means to repair, if one pin should break that would mark the end of our fun.

Fortunately, when we salvaged it from its previous home we noticed that those pins were quite strong in union. A single pin is fragile, but all twenty pins together was pretty strong. So we’ll solder up one of our cheap prototype boards to these pins just for the sake of holding them together. It was a bit tedious getting all the pins to line up. Two more boards were used as spacers to give us the desired distance.

Once lined up, the VFD was soldered in place.

As anticipated, the assembly is now much more robust. Now we can work with this module with a lot less trepidation: if we should break a pin now, the break should stop at this PCB where we could still recover, not break all the way to VFD glass where we couldn’t.

We still had a lot of usable pin length remaining, so we left this unionizing PCB alone and soldered another one where we’ll actually attach wires and headers.

While this soldering work was underway, a breadboard was populated to test our design for interfacing our driver PCB and the VFD. Each VFD grid and segment wire received a 10 kilo-ohm pull-up resistor to the 30V line, and path cords were installed to map pins from our generic driver board to this specific VFD as per the pinout diagram we drew up earlier. Now the VFD, with its new more robust pins, should plug right in.

Finally, power delivery which is far more complex than what a LED project requires. This initial integration test had three different power sources. The power transformer we salvaged alongside this VFD delivers (among other voltages) the ~2.5V AC we need for our filament. With a few rectifier circuits, it should be able to deliver the other DC voltages we need as well, but we weren’t going to worry about it until we got this test working. In the meantime, a HP inkjet printer power supply delivered the 30V DC for grids and segments, and a Raspberry Pi delivers the 5V logic power as well as I²C control signals.

We plugged everything in and… it lives again under our control!

Once I had assembled our first prototype vacuum fluorescent display (VFD) driver circuit board, it’s tempting to connect it right up to our salvaged VFD and bask in its glow. But that would have been premature: there were bugs still to be ironed out with potential for fatal errors to damage or destroy our one and only salvaged VFD.

The first tests were what we used to test our bread board prototype: drive a 4-digit 7-segment LED, a Lite-On LTC4627-JR from my earlier PIC display driver project. From there I could verify 8 of 11 segments on this driver board functioned as expected by a demo program that constantly flips between two patterns, one spells “Hello” and the other “Ahoy”.

I then tested the I²C capability with a small update, changing the second pattern from “Anoy” to a very mutilated “MLE”. (It’s hard to do “M” on 7 segments.)

These experiments exposed a few minor bugs, just as tests were supposed to. After they were fixed I expanded my test with a second LTC4627-JR for eight total segments. I also expanded my use of I²C by continuously updating all digits from Raspberry Pi to create a text marquee. This experiment exposed some timing issues that is visible as a slight lag just as the “u” in “thank you” came up on screen.

That lag was diagnosed to my code inside a timer interrupt service routine (ISR). On a simple chip like this PIC, a hardware interrupt signal could not interrupt another ISR already in progress. So when my code inside a timer ISR grew beyond a certain point, it started interfering with timely response to hardware interrupts necessary for I²C communication. The solution is to put my timer ISR on a severe diet, moving almost all of its bulk elsewhere so that I²C handlers could run with minimal interference.

Once the software problems were sorted out, I started experimenting with the flexibility given by this project’s design decision to keep PIC code simple. It meant I could play with patterns that have nothing to do with displaying letters or text. I decided to try a pattern that keeps a single lighted segment running through the display in a figure 8. When cycled through all digits like a text marquee, it gave us a neat looking wave. I look forward to more experiments in driving a multi-segmented display in unusual ways.

In addition to exploring the idea of aggressively packing chips densely, there were a few other design considerations in building our prototype driver PCB.

Why I²C?

When I was tearing my hair out trying to figure out why a tried-and-true piece of template code has stopped working, I was sorely tempted to move to some communication protocol other than I²C… good old TTL Serial always looks good whenever I run into a problem.

The answer is voltage compatibility: I want my driver board to be capable of running at either 3.3V or 5V, which is easy to do with an old 8-bit chip like the PIC16F18345 I’m using: give it power anywhere in its specified range (and probably a bit beyond that range) and it’ll happy run. But this doesn’t necessarily apply to whatever my driver board will accept commands from. Fortunately, I²C was designed from the start to accommodate devices that work off different VCC levels. So something like a Raspberry Pi (3.3V) can communicate with my PIC. (5V).

Pin Assignments

In addition to moving the decoder output pins so it could be placed immediately adjacent to my PIC, I also took advantage of my particular PIC’s feature of Peripheral Pin Select (PPS) by moving the I²C pins (handled by Master Synchronous Serial Peripheral – MSSP) from their default to RA4 & RA5 for easier routing on the PCB and easier data writes in my code.

Pin Isolation

This is the first time I planned to use pins RA0 and RA1 on a PIC16F18345 project. I usually avoid them because they are used as in-circuit system programming data (ICSPDAT) and clock (ICSPCLK) pins. They’re perfectly usable outside of this scenario, but they must be disconnected from the rest of the circuit during programming. In this case, it meant I had to put in two jumpers that are removed during programming. The jumpers are installed exactly where my PICkit 3 programmer would attach, so there’s no way for me to forget to remove them when reprogramming this chip.

Wires From CAT-5E Network Cable

Even with all these concessions to easier routing on a prototype PCB, there will still be an unavoidable nest of wires. In order to keep things tamed, I harvest my spool of CAT-5E network cable for wires that are thin and rigid. Thin wires help me fit more wires in a small volume, and rigid wires stay where I’ve put them. In addition to being affordable and easily available, the four twisted pairs of wires give me eight different insulation colors to help me keep track of which wire goes where.

While we were building a breadboard prototype for our VFD driver, we weren’t terribly concerned with chip layout as it was largely constrained by the practicalities of a bread board anyway. Once we start thinking about transferring it to a PCB, however, we had more flexibility to be creative. The prototype printed circuit board is a grid of through-holes that we can use in any way we like. How shall we abuse this power?

While looking at how wires were run on our bread board, I noticed that there were a few pins that were perfectly aligned from one chip to the next. Placing the decoder output on PIC pins RB4, RB5, and RB6 lines them up perfectly with the input pins of 74138 decoder. As for the decoder output, six out of eight pins were directly lined up with a ULN2003.

This allows some fairly straightforward wiring solder as these wires will not cross over each other and won’t tangle up and make a big mess. By itself that is valuable, but we were tempted to go one step further: how about we eliminate the wires entirely and jam those chips together? If they share the same PCB through-hole, that would eliminate wire soldering entirely.

It sounded good in theory, but in practice the chips are just a little too large for us to fit them three-abreast. I could push my decoder up against my PIC, but couldn’t push the Darlington array alongside the decoder as well. This is probably for the best – there’s value in having the ULN2003 lined up with its siblings.

We may learn in the near future why this is a bad idea, but it’s all part of the fun: trying things and seeing how they work. (Or not, as the case may be.)

Once we had some idea of what we wanted to do, it’s time to start wiring things up on a breadboard to see if it actually does anything interesting. Dealing with actual chips meant reading their data sheets and figure out where the rubber meets the road and how well theory meets practice.

Not having a lot of first-hand experience with such modules, it was a great way to learn by doing. The first surprise was behavior of the 74HC138 decoder module: Conceptually it takes a three-bit input and decodes it to one of eight pins. Conceptually we thought that meant raising one of eight pins high, but it actually lowers one of eight pins low.

We thought this was going to be a problem and started looking into inverters… before we realized it is a perfect pairing with ULN2003 to do what we actually wanted: The ULN2003 line inverts its input in the sense that when an input pin is high it connects the output pin to ground. So the output of a 74HC138 decoder (seven pins high, one pin low) driving ULN2003 results in seven pins connected to ground and one floating. We can work with this.

A 4-digit, 7-segment LED module stood in for the vacuum fluorescent display while all the logic got worked out. This flexibility is exactly where a breadboard is strong, letting us experiment and verify pieces of our circuit piecemeal. But it is not great for looking respectable or for long-term reliability, so once it successfully ran the LED module, we start planning to commit the design to soldering parts on a prototype PCB.

For the VFD driver project, there were software design motivations to keep things simple. But that’s not the whole picture, there were also motivation from hardware constraints too. My previous projects to make a PIC drive a multi-segmented LED display had fairly simple wiring that connected most output pins of my PIC to current-limiting resistors. A few of the lines could have seen current flow higher than what a PIC is capable of handling, and those were handled with simple transistors. I knew there existed chips designed specifically to drive LEDs, but I wanted to learn the principles of controlling one myself.

Building something to drive a VFD requires dealing with voltages different and sometimes far higher than what is required to drive LED display modules. During our probe of this specific VFD we saw 2.5V AC and 30V DC, atypical of logic circuits. And just as there existed dedicated LED driver chips, there exist chips specifically designed to drive VFD modules, but again the project goal was to learn by building one ourselves.

So we turn to our standard electronic hobbyist toolbox item for controlling power and voltage beyond what our standard parts can handle: the ULN2003A line of Darlington arrays. The go-to solution for controlling inductive loads like relays and small motors, it can handle voltages up to 50V which we need for a VFD.

And again, with multiple different display projects on the horizon, it didn’t make sense to create a controller with hardware pinout specifically tailored to a specific unit. To keep things simple and consistent across displays, all of our controller outputs will be either left floating or tied to ground. If a particular device desires a particular pin to be at a higher voltage, we’ll have to wire up a pull-up resistor on that device’s specific interface board. We will learn if this concession to consistency will cause problems down the line.