Amy Time! Nixie Tube Clock

There are few original ideas: we revel in our small successes, secure on the shoulders of generous giants. I hope the reader can use and improve upon the information and techniques I present below; most didn't originate with me. I've included a lot of excruciating detail for this reason. Design files are linked throughout the text.

This project started life as a cute General Electric battery charger I bought at a hamfest. There was no model number, and it wasn't obvious what the charger powered. The extruded aluminum enclosure looked perfect for a clock.

Many years ago my brother gave me a nixie tube clock kit from Peter Jensen's TubeClock.com. The kit was easy to assemble and has worked very well. I was impressed to see that instead of rare and fragile 74141 nixie driver chips, or discrete transistor drivers, Peter had used a HV5622 high-voltage shift register manufactured by Supertex Inc. (now Microchip). I hadn't encountered the part before, but I saw that this approach conserved PCB real estate; required a minimum of microcontroller I/O; and relied on a relatively cheap current production part.

For my clock, I intended to use the same approach. I fabricated a couple of breakout boards for the chip to cultivate familiarity with its operation (and maybe reuse for future, less space-constrained nixie projects).

I wanted an efficient, reliable power supply based on the common and inexpensive MC34063. Threeneuron's design seemed well thought out, and he included part numbers for the inductor and other critical components. I built a couple of power supplies based on his circuit. They perform well, but the abrupt off transition of the power FET results in dramatic high-frequency ringing. The switching frequency is right at 1 kHz for a 200VDC output, and the ringing sweeps across 458-543 kHz with a maximum amplitude of 102VAC (+86V/-16V). It's obvious that a snubber network is needed, but I'm not sure how to accurately measure the total capacitance at the FET output or calculate the appropriate values. If you, the reader, can offer any tips, I'd love to hear from you at the email address below.

PSU efficiency is very good, around 70-80%. It also "sings" audibly under load, probably from electromechanical vibration in the power inductor at the switching fundamental. I would be curious to see if a snubber network, or better output decoupling, would help this problem. Since I was working on a short timetable, I accepted the basic success and moved on.

nixie-hvps.pdf	HVPS schematic diagram in PDF format
nixie-hvps.png	HVPS schematic diagram in PNG format
nixie-hvps.zip	HVPS schematic and PCB layout in Eagle 5.12 format, and PDF schematic

I purchased two lots of 6 Russian surplus IN-12 tubes through eBay, one with bakelite sockets. Shipping time was about three weeks. Quality wasn't on par with the US-made tubes I've seen; there's a lot of dimensional variance in the glass envelope and it's sometimes crooked relative to the internal structure. If you're building a clock using IN-12s, I'd order extras and select for best appearance.

IN-12.pdf

Russian-language IN-12 datasheet

It's important to carefully understand the power requirements of your tubes. I started testing with a 20K anode resistor at 200V, which seemed reasonable from the information on Threeneuron's site. After running a tube for several hours, I could see evident cathode poisoning. The anode current was 3.5mA, right at the top of the range for the IN-12. After some experimentation, I settled on 53.4K (two 26.7K resistors in series, since that's what I had). This supplied a much more conservative 1.3mA, which should promote longevity of the tube and still yields decent brightness and visibility.

With the prototype hardware ready, it was time to start writing code. The HV5622 is easy to address in the typical manner of a shift register. For nixie tube applications, Polarity (_POL, pin 27 of the PQFP variant) should be permanently wired high. The HV5622 appears to power on with some or all of the outputs enabled, possibly in a random state. It's wise to start with Blanking (_BL, pin 33) low, shift in your desired initialization state, then bring Blanking high to enable the display. (Polarity and Blanking have very similar functions, which could probably be inverted. The TubeClock.com kit ties Polarity high, so that's what I did.)

Pull Clock high (CLK, pin 28).
Set Data In (pin 32) high or low as needed.
Pull Clock low. This shifts the bit into the register.
Repeat until all 32 bits are shifted in.
Pulse Latch Enable (_LE, pin 31) high then low. The data is copied to the output latch.

Only 28 of the 32 outputs are needed for a 4-digit clock, 27 if you're satisfied with 12-hour time: 1[-2], 0-9, 0-5, 0-9. It happened that I needed to address the outputs in counterclockwise order; the shift routine is easily modified for big- or little-endian shifting.

time.asm

AVR assembly language firmware for ATmega48, compiles with gavrasm

Below are pictures of the test setup. I used a four-channel Tektronix 2245A, triggered off Latch Enable, to view continuous data sent to the shift register and help debug my code. There's a triggering delay, so the signal wraps around the display. The four peaks correspond to four active outputs, one for a numeral in each tube.

With the code more-or-less squared away, I turned my attention to mechanical design. Using digital calipers I measured the chassis, the tubes, and the sockets, and made 2D vector drawings in Inkscape. I scanned the original faceplate and imported the resulting bitmap into Inkscape as well, scaling the object dimensions to match width and height measurements of the real piece before converting the outline to a vector object with opaque fill. Using these elements I was able to choose the best placement for the tubes in the chassis and mock up a dummy faceplate before doing any machining. I printed out paper templates and affixed them to the metal parts with tape, then used a center punch to define drill holes. This resulted in pretty good accuracy: the final clock assembly matched my drawing in Inkscape to within about 1/32". The tube sockets just barely fit in the steel chassis, and the faceplate needed to be positioned just right to clear the enclosure--working from dimensioned templates was tremendously helpful, and probably prevented a lot of rework and misery.

IN-12.svg

Dimensional drawing for IN-12 tube, socket, and front panel cutout, in Scalable Vector Graphics format

Pictured below are the original chassis with added framework for the tube sockets, and the dummy faceplate. Looks encouraging!

Painting the enclosure. The original navy blue enamel wasn't without its charm, but it had undergone a lot of abuse and I wanted something more festive. I used Testors Model Master spray paint. This was a terrifying experience for a novice. The paint contains a lot of solvents which take a while to evaporate, so it stays liquid for several minutes. Fresh from the can it splatters and sags; over time surface tension draws the paint into a beautiful smooth finish that conforms to the texture of the original enamel while concealing small blemishes. I applied a cream coat over the enclosure components, let it cure for 48 hours as recommended, and added red detailing using masking tape and waxed paper. I had hoped the waxed paper would mask without sticking to the cream, but it was a terrible choice: the paint solvent attacked the "wax" coating and red paint bled through in spots. The finish was very difficult to touch up without leaving visible evidence.

After several days the paint had outgassed enough to be reasonably durable. It's not as tough as powder coating, but it's not soft or pliable either.

Time to get serious. I had idly sketched several ideas for the faceplate legend in CadSoft EAGLE, since the grid feature made it easy to lay out the (mid-century) modern, rectilinear design I had in mind. (Inkscape has a grid as well, but Eagle was open at the time.) The doodle turned out well enough that I exported it in PDF format, imported to Inkscape, and converted to paths. Eagle is perfectly usable as a simple vector-drawing program; I've used it to make several front panels, lettering and all. (Here's an example, in the form of a moderately unsuccessful experiment to engrave aluminum.)

I finalized the faceplate design in Inkscape and printed it in mirror-image on the photo paper I use for PCB toner transfer. (Fry's Great Quality brand, sadly long discontinued. Some day I'll exhaust my supply and have to find an alternative.) I cut stainless steel sheet to outsize dimensions, brushed it with 600-grit sandpaper to promote toner adhesion, and ironed the transfer on high for several minutes. The results were very good, better than I typically see with copper: only a few specks of toner, and one outside edge of the design, failed to transfer.

On a whim, I painted over the transferred toner with a fine-tip paint pen. This seemed unnecessary at the time, but turned out to be a prudent precaution. I wanted a deep relief in the steel, and etched the workpiece in room-temperature ferric chloride for around 40 minutes. The toner layer is slightly porous; this doesn't matter for thin PCB copper cladding, but after the longer etch time discoloration and pitting were visible wherever I hadn't touched up with paint. Using the paint pen, I filled in the relief and let dry.

As I'd discovered previously, freehand abrasive brushing can offer a pleasing finish; but it wasn't straight or consistent enough to reproduce the industrial brushed look I wanted. I affixed 220-grit sandpaper to one 2x4 and the workpiece to another, using fiberglass carpet tape. Using smooth, firm strokes I slid the 2x4s past each other, keeping them parallel against the floor. Stopping or starting mid-stroke left unwanted marks, so finding the right technique took some experimentation. Excess paint and much of the pitting was removed; the resulting finish was excellent.

I expected the workpiece to warp and burr slightly during the shaping process. This would cause the sandpaper to ride high in some spots, resulting in an uneven finish, so I elected to brush first and handle very carefully. Over the course of shaping, the faceplate did indeed incur some small scratches and burnish marks. I was able to remove these with a tab of sandpaper taped to my finger, and light straight strokes over the trouble spots.

I had chosen a 1/4" radius cutout for the rounded corner of the IN-12, adding pilot hole marks to the toner transfer design in Inkscape. I used a step drill to drill these out to 1/2", and also drilled and deburred the holes for the LED seconds display. Using a Klein 76011B nibbling tool, I roughed out the center portion of the tube cutouts. The Klein nibbler is a terrific tool, once you remove the hold-down clamp that obscures the cut and mars the workpiece. Oil it when it squeaks, and it will provide many years of service. Nibbling does distort the steel, so I had to maintain at least 1/16" of distance from the edge of the cutout. I fixed the faceplate in a vise with chunks of ABS plastic for makeshift soft jaws, and painstakingly filed out the remaining material along the guide lines etched into the workpiece.

In retrospect I wish I'd spent a little more time finishing out the cutouts, maybe with fine-grit sandpaper wrapped around a dowel. I was afraid of removing too much material, or slipping and marring the finish, which is why I used hand tools exclusively. The result still looks fine from a normal viewing distance, and pretty good close up. It's hard to know when to say when.

I clamped the increasingly delicate faceplate between two blocks of wood to prevent warping or twisting, using waxed paper as a medium to prevent the wood fiber from marring the finish under vibration and pressure. I used a 1" belt sander I bought at a hamfest, originally built for shaping golf clubs, to grind away excess material around the faceplate outline, using 80-120 grit belts for rapid removal and a 400-grit belt for finishing. This was an effective and controllable process; but friction increased as grit was stripped from the belts, heating the workpiece and discoloring the edges. I started quenching with water after each pass, and this was partially effective.

The heat caused the waxed paper coating to melt and stick to the faceplate. It had to be very carefully removed with acetone. Over the course of this project, waxed paper is trending as a candidate for worst craft material of all time.

The circuit design and PCB layout happened concurrently with chassis fabrication. It wasn't until this point that I had a concrete idea of how everything could be positioned in the cramped enclosure: power entry, time set controls, display wiring, etc. The circuit comprised the following elements:

Maxim Integrated DS32kHz 32.768kHz TCXO, which claims +/-1 minute/year accuracy at typical room temperature and comes in a convenient SOIC package.
Atmel ATmega48 MCU, clocked from the internal RC oscillator and using the DS32kHz as an external timer interrupt source.
An array of 6 LEDs for a seconds display. I had several ideas for various nifty time visualizations: PWM heartbeat, bargraph, Larson Scanner, AM/PM status bits, etc. The timer code counts in half-seconds to enable per-second blinking, which I didn't end up using. In the end I fell back on the simple binary counter placeholder I wrote for debugging the code, preset to start at 4 and roll over at 64 so all six LEDs would indicate the last second of the minute. This doesn't display the current second in a meaningful way, but it does indicate the passage of time in a pleasant and engaging format.
Buttons triggering external interrupts for setting the minute and hour. I generally use external pull-ups because AVR internal pull-up resistors aren't actually a guaranteed value, and in fact vary from pin to pin. For the physical buttons, I used tactile switches with stainless steel acorn nuts as button caps. The nuts were drilled out to a precise diameter and depth to fit over the switch actuators, and were held captive by holes in the rear panel.
A simple 5V linear power supply for the logic, and the high voltage power supply design I'd prototyped. These are both powered from a 12VDC SMPS wall wart, but the circuit as designed should be able to tolerate input voltages from 7-24VDC and includes reverse voltage protection.

The cathode connections on the board are separated into discrete groups for each digit, in an effort to prevent coupling from one digit to an adjacent active ground for the next digit. This shouldn't be a problem at DC, but the HVPS filtering isn't perfect and I was concerned that any AC component might couple current and cause unwanted display artifacts. This was probably completely unnecessary. It would be interesting to connect a set of tubes in a sub-optimal way, e.g. over a few feet of fine-pitch ribbon cable, to test whether this concern is warranted.

After the board was fabricated, I ended up doing some hackwork to add two 26.7K resistors in series for each anode. Additionally, the prototype PSUs used 2.2uF 250V MLCC capacitors; I substituted aluminum electrolytic out of concern for MLCC reliability. Since the tab of the IRF740 is at HV potential and would contact the case for thermal dissipation, I used the isolated-tab IRFI740 variant in the clock. A mica insulator might allow better thermal transfer, but I wanted a high margin of safety. The tab of the mounted IRFI740 was barely warm to the touch in operation.

The ISP header was originally intended to extend from the top layer of the PCB. After fabricating the board, I decided to place it on the other side and add an access plug in the rear panel. This allowed for uploading new firmware without disassembling the clock, but the pinout is mirror-image: 2-1-4-3-6-5. Alas! I used three 2-wire jumpers to swap the signals.

time-schematic.pdf	Schematic diagram in PDF format
time-schematic.png	Schematic diagram in PNG format
time.zip	Schematic and PCB layout in Eagle 5.12 format, PDF schematic, and code.

With the PCB mounted in the chassis, I wired up the tube sockets using tinned solid hookup wire and teflon sleeving. The wires were routed to avoid any contact and possible coupling between digits (again, probably overkill). I opted to buy sockets with the tubes and wire them point-to-point, reasoning that this would be faster and simpler than building sockets from scratch on a custom PCB. (See Notes on DIY Nixie Tube Sockets.) In retrospect, the custom PCB route might have been cheaper while facilitating assembly considerably. Socketed mounting did offer a little bit of flexibility over tube position. All the IN-12 sockets I have are slightly rotated relative to the tube. Additionally, poor QC as noted above means tube geometry varies noticeably. It was useful to be able to tweak each tube position for best appearance before tightening the mounting screws.

And, it's alive! Everything worked great from first light, except that I'd wired both numerals 1 and 2 in the first tube to give the option of 24-hour time, and my code somehow transposed them. A small software fix, plus a few tweaks for button debouncing and the LED display, and the software was ready for a few days of timekeeping and burn-in.

The clock draws between 80-140mA at 12V, depending on how many LEDs are lit.

I did some deep and lengthy head-scratching over how to fasten the faceplates. The original battery charger used panel-mount components to hold the front and rear panels to the chassis. I didn't have anything like that here, and I didn't want to use visible fasteners, but the panels needed to be secure. I ended up fabricating brackets and epoxying them to the panels with JB Weld. I don't have a lot of confidence in JB Weld for small scale applications, but it worked great here.

Happy Birthday, Amy!

It's been a few weeks (update: five years), and the clock keeps time as well as the other devices I have to compare.

If you have any questions or comments, or if you've derived any use from the information presented herein, I'd be delighted to hear from you.

Contact: reboots at g-cipher.net XHTML and CSS compliant :wq