For the past few months (feb. 2016) and over around 200 hours I have been designing and building a nixie-based alarm clock. The inspiration came from those wake-up light clocks that slowly brighten a powerful light source before waking you up in order to trick the brain into thinking it's morning. That was early September 2015, and I think it's safe to say that the project escalated quite a bit. When I miraculously salvaged four Amperex ZM1000 nixie tubes in November from a dumpster and managed to bring them to life on a breadboard, I decided to really go all-in on this project. This has been the first large-scale project I have done in electronics, and will surely not be my last.

The clock now features an 8 digit, 7 segment display and a rather expansive menu system, allowing the user to change just about any thinkable parameter in the system. Interaction with the menu system is done with a rather lovely 12 step rotary encoder, and arming the clock is done by an arm switch and key. The transparent plexi glass front allows insight into the guts of the system.

I will present each subsystem of the clock in the order of which they were designed and implemented, thus giving the reader a feel for the design process. Feel free to use any idea, code, schematic or board layout, but remember to reference this website or myself if you do.

The clock in all its glory glory!

I decided to start off by designing the 7 segment display, imagining it would come in handy for debugging. Being a fan of everything retro, I fell in love with the Bubble Display from Sparkfun. These were normally seen in calculators and such back in the day, and are nothing short of awesome. While waiting for those to arrive, I set about designing the support circuitry for the 2x4 segments. This particular display is common cathode, meaning that every segment in the digits has common ground. There being eight digits, I eventually decided to use two HC74595 shift registers to drive the anodes and one TPIC6B595 shift register to sink the current from each cathode. I could have used one single shift register for all the anodes, but that would result in a 1/8th duty cycle, which looked a bit dim during breadboard tests.

I was planning to use the laser print method to create the PCB. Despite designing a two layer board, I struggled to organize all the connections in order to keep the shift registers and the displays on the same board and still keep the module's size reasonably small. I therefore decided to re-organize the connections with ribbon cables from the shift register PCB to a perfboard onto which the display was soldered. This solution became problematic when the whole system was being assembled, since a thick bulk of ribbon cables is rather unwieldy. Read the paragraph on Physical assembly to see how I turned that problem to my advantage. You can find a download link for the EAGLE files at the bottom of the page.

This is how I ended up multiplexing the display:

Despite a mere 25 % duty cycle and below-recommended segment current, the display is more than bright enough.

The screen assembly - before I noticed I had rotated the cathode wires on the right-most array.

Initial testing of common anode 7 segment displays. It's amazing how many jumper cables it takes!

After reading up a bit on wake-up lights and following a never-ending trail of Wikipedia articles regarding sleep, hormones, light, color rendition indeces, sunlight and how humans perceive light, I realized why those commercially available wake-up lights are so expensive. From what I could gather it seems that the effect is strongly dependent upon the quality of light, which can be quantized as the light source's Color Rendition Index (CRI). CRI is a number which in essence tells you how well a particular light-source is at revealing an objects true color in comparison to a natural or ideal light source (often referenced to sunlight). Basically: High CRI = good light. The idea behind the wake-up lights is to trick your brain that morning is coming by simulating a sunrise, and is backed by research:

Dawn simulation units mimic the gradual twilight transitions found outdoors during spring and summer. These light signals are presented to people while they are still asleep, and while their eyes are still adapted to the dark of the night. This is the point where the circadian system is most susceptible to the phase advances of gradually brightening light. The presentation of diffuse, broad-field illumination evokes an antidepressant response and helps those affected by seasonal sleep disorders to wake up normally.

Terman, M., & Terman, J.S., 2005

I decided to use FlexFireLEDs's ultrabright series LED strip for this project. They are indeed rather expensive, but in turn they have a guarateed CRI of 93+, a market best as far as I can tell. They are also very bright at around 450 lm / 5W per foot. By comparison, a standard 40 W incandescent bulb is 450 lm omnidirectional. I can't look directly into the lights at full brightness without discomfort.

These LED strips run at 12 V, so by this stage in the project I had decided to use a standard 12 V power brick to drive the system. Driving them is easy; A simple, generic power MOSFET connected to an 8 bit PWM pin on the microcontroller enabled me to dim the lights.

The LED strip running at around 5 % duty cycle. It's painful to look directly into the light at 100 %! Notice how the first three LED's are dead...

I could not for the life of me get the solder to reflow on the top of the pads, so I soldered from underneath. The heat was probably set too high, so the first row of three LED's died. This is not noticeable though -- see the section on Physical assembly.

A big thanks to for providing the ancient datasheet for these nixie tubes!

This part of the project that had me worried for a while. The four ZM1000 nixie tubes I salvaged require a minimum of 170 V DC supply voltage for ignition and an average anode current of 2.5 mA. Given the extreme rarity of this particular nixie tube (and, surprisingly, the even rarer sockets), I was very afraid I might damage them by misunderstanding their parameters. Reading the datasheet carefully, I decided to leave the primer pin floating, and supplying the anodes with 180 V DC through a 20k reistor. The voltage source is a cheap 180 V boost converter off eBay, but in hindsight I wish I had built my own, as the one from eBay has a terrible coil-whine. Wrapping it in some cloth seems to kill most of the whine until I get around to fixing it more permanently.

Now, sinking current at those voltages is not something your ordinary 74-series shift register can hope handle. Luckily, you can still obtain IC's from the 1970's through eBay which do just that. I ordered SN74141-equivalent chips from Ukraine: MilSpec TESLA MH74141, and they work a charm. The following is the schematic I ended up with.

The 74141 chips are BCD, which means that every nixie tube needs its own chip. Every chip has four data inputs; A, B, C and D. This adds up nicely so that four digits can be controlled with two shift registers. Four bits give 24 = 16 possible values, so binary inputs of 10-15 are invalid inputs and will result in all the outputs being high impedance -- perfect for blanking individual nixies.

Finding an EAGLE library part for them was impossible, so I designed a package for them. You can fetch the schematic, board layout and the library part from the bottom of this page. In order to pay the respect these beautiful things deserve, I decided to have the circuit board manufactured. Feeling confident after successfully designing the display driver board, I set about designing this one too. As with the rest of the system, I made sure this board too was modular, with the following header inputs: 180 V, GND and 5 V plus DATA, LATCH, CLOCK and DECIMAL divided over two ribbon cables. The DECIMAL input controls the gate of a single high-voltage NPN transistor (MPSA42) which sinks current from the third nixie's decimal point cathode, thus enabling a blinking decimal.

ZM1000 nixies running at 180 V DC, 2.1 mA with a 20k anode resistor. I won't say more -- just look at the sexy things.

A closer look at the nixie driver. I am using header pins and ribbon cable as connectors -- cheap and easy, though unreliable. These will be sealed with heat shrink tubing later.

Bottom of the circuit board

Top of the circuit board

Anticipating the everlasting work-in-progress-nature of this project, I went for perfboard for the main circuit board. Connectors are rather expensive, and I didn't bother ordering them through eBay. Most of the board-to-board connections are therefore done by soldering female headers to ribbon cable and mating the header with a male counterpart on the PCB. This isn't the most robust solution, but if you make sure the solder joints are good and you add heat shrink tubing to the connectors, it seems to work okay for low voltage connections. I did go for proper JST connections on the power connectors, which also ensures I don't connect 12 V or 180 V backwards to ground or 5 V. This did happen on one occasion, which resulted in a lot of magic smoke and the demise of an ATmega chip and several innocent shift registers. Luckily no critical components were damaged.

The timekeeping is done with a breakout board based on the very accurate DS3231 RTC module with its own dedicated button cell battery. The system is powered by 12 V from a (rather large) random power brick I had laying around.

Layout of the main perfboard.

Interrupts and timekeeping

At first I wrote all the code in Visual Studio with the Visual Micro plugin for an Arduino Uno, but I soon switched platforms to the ATmega644P, as I was interested in going deeper into microcontroller programming. This part of the project has definitely been the most time-consuming, mostly due to the fact that I had never programmed anything AVR, apart from Arduino. It was a steep learning curve, but I am glad I didn't give up along the way. The code definitely shows that I am new to this game, but at least it works! I won't go through every line (if you have any questions or feedback, feel free to contact me), but I'll explain key features.

Everything in the code is interrupt-driven. On the ATmega644p there are three timers; two 8 bit timers and one 16 bit timer. The 16 bit timer is used in CTC mode to interrupt every second, calling an ISR every time the counter reaches the preset value in the ORC1A register. I derived the following formula from the datasheet to calculate the overflow compare value:

\text{OCR1A} = \frac{f_\text{clk}}{prescaler \cdot f_\text{ISR}}

where fclk is the system clock frequency (16 MHz), N is the prescaler value (1024) and fISR is the interrupt frequency (1 Hz). For these values I get OCR1A = 15625. The theory behind timer interrupts in AVR has been described exhaustively in numerous tutorials, blogs and posts around the web, so I will not go into detail about it here.

Display code

In the file display.cpp, you'll see that the class Display has a member function, refresh(), which is called every 1/100th second by another timer interrupt. This function cycles the displayed character as illustrated above. The function updateString(char* newStr) generates a string of bytes which is compatible with the configuration of shift registers based on the c-string newStr and a long lookup-table found in header.h.

Nixie digit code

The code which handles the nixie tubes resembles the one controlling the display, but with the added need for decimal - BCD conversion. Every 15 seconds the microcontroller fetched the current time from a DS3231 RTC module. If one minute has passed, the function Clock::refresh() takes the global variables seconds, minutes and hour and generates BCD values for the shift registers. It also accounts for whether or not the user wants to print leading zeros and if he/she wants the decimal point to flash. All this is done on lines 30 and 31 in Clock::refresh() by the two ugliest lines of code I have ever written -- but they work. Note to self: don't code after 3 a.m.

Rotary encoder: decoding and debouncing

Quadrature rotary encoders are knobs you can turn infinitely, as opposed to a potentiometer. They are also digital by by nature, and are therefore very well suited for simple menu interactions and so on. After studying Nich Fugal's very detailed walk-through on how they work and how you can use them with an Arduino, the setup was easy. The encoder I use also incorporates a push-button in the encoder shaft. In encoder.cpp you'll find the function Encoder::readEncoderPosition() which is adapted from Nich Fugal's blog post. This function both debounces the encoder inputs A and B and calculates the probable direction in which it was turned.

Pulse-width modulation of the LED strip

Timer0 of the ATmega was configured for fast-PWM and toggles PB3 (OC0A) on compare match with the OCR0A register. Thus, by changing the value of OCR0A, we change the duty cycle of the PWM signal which drives the LED strip. As you might know, humans perceive the world mostly on a logarithmic scale, while microcontrollers are most definitely linear in every digital aspect. To compensate for this I decided to let the user choose between an exponential and a linear basis for dimming the light. The function Light::initializeLists() generates two lists containing respectively linear PWM values and exponentially spaced PWM values. Unfortunately, due to the properties of exponentials and the 8 bit limitation of OCR0A, the latter list contains a mere 41 steps. This results in a rather jagged dimming of the light, but at least it looks linear. I just occurred to me, though, that a sunrise might follow neither of these patterns. I promise to look into which patterns produce the best effect and post my findings.

On a side-note: Inducing lucid dreams!

Having been intrigued by the concept of lucid dreaming for a long time, I now finally have equipment which enables me to induce a lucid dreamstate. Research does back this method of inducing a lucid state. In one study, flashing lights were administered through goggles worn by the sleeper, inducing lucidity 33 times over a total of 58 nights with 44 subjects (LaBerge, Levitan, Rich, & Dement, 1988). Oh yes.

This is not something I have been able to refine yet, though. What separates my crude experiments from the study is that they had sensors detecting eye movement through the sleeper's eyelids and activated the flashing lights only during REM sleep. I am still only smashing rocks together, but I have had promising results. For a couple of weeks I have set the alarm clock to flash the LED strip at around 05.00 a.m. On at least two occasions the flashing has resulted in terrifying nightmares in which I was sure the world was ending and I was going to die (the lights being accompanied by thunderous noises, interestingly), though not resulting in a lucid state as far as I can recall. I will play around with the light intensity, the flashing frequency and the time interval during which the lights flash and post my findings.

Update: I am designing a wearable system which, using electrooculography, will be able to detect REM sleep. Through the use of light stimuli provided by the wearable, I will hopefully be able to induce lucid dreams. Click here to read more!

At this point I'm several months into the project and I am very anxious to collect everything and install it in a permanent box. Having dry-fitted the system in a suitable cardboard box for longer than I care to admit, I went down to my local hardware store and purchased a length of solid oak shelving (15 by 3 cm), an acrylic sheet, decorative molding, some tools, miscellaneous screws, angle brackets and so on. Loosely following the dimensions of the cardboard box I constructed a basic shape. The top plate is screwed in from the sides, and the plexiglass front can slide out.

I drilled undersized holes in the back plate and hammered in PCB spacers in order to fasten the three circuit boards, which is surprisingly sturdy. The arm switch and the rotary encoder are fastened with nuts through the plexiglass. Fastening the LED strips to an angled wooden molding which spans the entire width of the box I keep the inside of the box shielded from the light as well as directing the light outwards. Using very fine sandpaper I diffused the bottom of the plexiglass in order to hide the LED strip as well as spreading the light. This worked really well; It hides the fact that the light consists of several small LED modules without compromising brightness or light quality. It is, however, important to polish vertically, not horizontally. If the minute grooves in the acrylic material from the sanding are parallel to the LED strip, the resulting effect will not be diffusion -- rather, you will reinforce the feeling of the light being "one-dimensional".

Having designed this system without the assembly in mind, everything was a mess of ribbon cables. The highly modular nature of the system is one of the reasons why the box is so large. In order to clean up the mess of cables I stripped some thick copper wire and shaped it into a hook. The shape became quite rigid after applying loads of solder. Fastening it to the bottom plate of the box I was able to secure the display right behind the plexiglass as you can see in the image to the right. Not only does this look cool, but it allows much more insight behind the display than what you would have if the display was fastened to a large PCB. I made another hook for cable management inside the box.

Final polishing was done with coarse sand paper and extra virgin olive oil. I hope my clock won't go rancid on me.

Dryfitting the system in a cardboard box.

A look through the transparent front. You can see the how the plexiglass is polished: Vertically, not horizontally