μWWVB is a watch stand that automatically sets the time on atomic wristwatches where regular WWVB signal isn’t available. The system acquires the correct time via GPS and sets radio-controlled clocks by emulating the amplitude-modulated WWVB time signal.

Most so-called atomic clocks aren’t true atomic clocks; rather, they are radio-controlled clocks that are synchronized to true atomic clocks. Radio clocks maintain time by using an internal quartz crystal oscillator and periodically synchronizing with an atomic clock radio signal. Quartz clocks have a fractional inaccuracy $\delta f / f \approx 6 \times 10^{-6}$, which means that they can gain or lose about 15 seconds every month. Official NIST US time is kept by an ensemble of cesium fountain atomic clocks — their newest clock, NIST-F2, has a fractional inaccuracy $\delta f / f < 1 \times 10^{-16}$, meaning that the clock would neither gain nor lose one second in about 300 million years.

Most radio-controlled clocks in the United States are synchronized to the WWVB radio station, which continuously broadcasts official NIST US time. WWVB broadcasts from Fort Collins, Colorado, using a two-transmitter system with an effective radiated power of 70 kW. Theoretically, during good atmospheric conditions, the signal should cover the continental United States. Unfortunately, I can’t get my wristwatch to receive the 60 kHz amplitude-modulated time signal in my dorm room in Cambridge, Massachusetts.

Taking into account frequency uncertainty, WWVB can provide time with an accuracy of about 100 microseconds. In the absence of WWVB, there are other sources that can provide reasonably accurate time. The Network Time Protocol (NTP), which operates over the Internet, can provide time with an accuracy of about 1 millisecond. GPS can theoretically provide time with an accuracy of tens of nanoseconds. I decided to use GPS, mostly because I didn’t want to make my WWVB emulator dependent on an Internet connection.

Building a WWVB emulator involves transmitting on 60 kHz. In general, it’s not legal to broadcast on arbitrary frequencies at an arbitrary transmit power, because transmissions cause interference. Many parts of the radio spectrum are already in use, as allocated by the Federal Communications Commission (FCC).

Luckily, the FCC grants exemptions for certain unlicensed transmissions, as specified by 47 CFR 15. This is explained in some detail in “Understanding the FCC Regulations for Low-Power Non-Licensed Transmitters”.

Transmitters in the 60 kHz band are allowed, and the emission limit at that frequency is given in 47 CFR 15.209. As long as the field strength is under $40 \text{ $\mu$V/m}$ as measured at 300 meters, it’s fine. In my use case, I have the transmitter within a couple inches of the receiver in my wristwatch, so I don’t need to transmit at a high power.

I designed and fabricated a tiny custom board designed to interface with a GPS and an antenna:

The board is powered by a $1 ATtiny44A microcontroller. I used a 20 MHz external crystal oscillator for the microcontroller so I’d have a more accurate clock than I would with the internal RC oscillator. The board has a Mini-USB connector for power, an AVR ISP header for programming the microcontroller, and a JST-SH 6 pin connector for the GPS. I included pin headers for the antenna, making sure to connect them to a port that works with fast PWM. I also included 3 LEDs as status indicators — a red LED for power, a green LED to indicate a GPS lock, and a blue LED to show the unmodulated WWVB signal.

I designed the board using the EAGLE PCB design software and milled the board from a single-sided FR-1 circuit board blank on an Othermill v2:

Once the board was finished, I used solder paste and a hot air gun to solder my components. Hand soldering surface-mount components is pretty painful, but using solder paste, the entire soldering process took only ten minutes.

For my GPS module, I used a USGlobalSat EM-506, a high-sensitivity GPS powered by the SiRFstarIV chipset.

The 60 kHz WWVB signal has a very long wavelength: $\lambda = c / f$, so the wavelength is approximately $(3 \times 10^8 \text{ m/s}) / 60 \text{ kHz} = 5000 \text{ m}$. It’s challenging to design good antennas for such long wavelengths — a quarter-wavelength antenna would be about 1250 meters long! WWVB uses a sophisticated antenna setup that’s automatically tuned using a computer to achieve an efficiency of about 70%. Luckily, for my use case, I didn’t need to worry about designing a really efficient antenna and doing careful impedance matching — I was transmitting over such a small distance that efficiency didn’t matter too much.

I didn’t want to build my own antenna, so I gutted a radio clock and repurposed its ferrite core loopstick antenna. Thanks to antenna reciprocity, which says that the receive and transmit properties of an antenna are identical, I knew that this should work.

I wrote software to periodically get accurate time via GPS and continuously rebroadcast the time following the WWVB protocol. The software is written in plain C and doesn’t use any libraries or anything. I used the CrossPack development environment on macOS for compiling my code and flashing my microcontroller.

Getting the software to work just right took a good amount of effort. To make it easier, I initially designed each component separately, and still, I ended up spending a lot of time debugging:

According to the datasheet, the EM-506 has a UART interface and supports both the SiRF Binary protocol and the NMEA protocol. NMEA 0183 is a standardized ASCII-based protocol, so I opted to use that over SiRF Binary.

After implementing software UART on the ATtiny44A, getting time data from the GPS was as simple as sending over a command to query for the ZDA (date and time) NMEA message:

$PSRF103,08,01,00,01*2D

In response, I’d get back a message with the current date and time (in UTC). For example, for 26 December 2016, 18:00:00, I’d get the following NMEA message¹:

$GPZDA,180000,26,12,2016,,*43

It was easy to parse the ZDA information to get the current date and time. However, the WWVB protocol required some extra date/time information not directly available in the ZDA data, so I had to write some date/time conversion utilities.

Leap year calculation was simple, and calculating the day of year was also straightforward.

Calculating whether daylight savings time was in effect took a little bit more effort. In the process of implementing it, I learned of a neat way to calculate the day of the week given the month, day, and year:

int day_of_week(long day, long month, long year) {
    // via https://en.wikipedia.org/wiki/Julian_day
    long a = (14 - month) / 12;
    long y = year + 4800 - a;
    long m = month + 12 * a - 3;
    long jdn = day + (153 * m + 2) / 5 + 365 * y +
        (y / 4) - (y / 100) + (y / 400) - 32045;

    return (jdn + 1) % 7;
}

int is_daylight_savings_time(int day, int month, int year) {
    // according to NIST
    // begins at 2:00 a.m. on the second Sunday of March
    // ends at 2:00 a.m. on the first Sunday of November

    if (month <= 2 || 12 <= month) return 0;
    if (4 <= month && month <= 10) return 1;

    // only march and november left
    int dow = day_of_week(day, month, year);
    if (month == 3) {
        return (day - dow > 7);
    } else {
        // month == 11
        return (day - dow <= 0);
    }
}

WWVB uses amplitude modulation of a 60 kHz carrier to transmit data at a rate of 1 bit per second, sending a full frame every minute. Every second, WWVB transmits a marker, a zero bit, or a one bit. A marker is sent by reducing the power of the carrier for 0.8 seconds and then restoring the power of the carrier for the remaining 0.2 seconds. A zero is sent by reducing the power of the carrier for 0.2 seconds, and a one is sent by reducing power for 0.5 seconds.

Here is the format of the WWVB time code, as documented by NIST:

I made use of the hardware PWM built into the ATtiny44A to generate and modulate the 60 kHz carrier for emulating WWVB. Working out exactly how to configure the microcontroller required careful reading of the section in the datasheet on fast PWM.

I used the following code to set up the 16-bit timer/counter:

// set system clock prescaler to /1
CLKPR = (1 << CLKPCE);
CLKPR = (0 << CLKPS3) | (0 << CLKPS2) | (0 << CLKPS1) | (0 << CLKPS0);

// initialize non-inverting fast PWM on OC1B (PA5)
// count from BOTTOM to ICR1 (mode 14), using /1 prescaler
TCCR1A = (1 << COM1B1) | (0 << COM1B0) | (1 << WGM11) | (0 << WGM10);
TCCR1B = (1 << WGM13) | (1 << WGM12) | (0 << CS12) | (0 << CS11) | (1 << CS10);
// fast PWM:
// f = f_clk / (N * (1 + TOP)), where N is the prescaler divider
// we have f_clk = 20 MHz
// for f = 60 kHz, we want N * (1 + TOP) = 333.3
// we're using a prescaler of 1, so we want ICR1 = TOP = 332
// this gives an f = 60.06 kHz
// we can use OCR1B to set duty cycle (a fraction of ICR1)
ICR1 = 332;
OCR1B = 0; // by default, have a low output
DDRA |= (1 << PA5); // set PA5 to an output port

After this setup, I could modulate the carrier by setting OCR1B. Setting OCR1B = 166 made a 50% duty cycle 60 kHz square wave, and setting OCR1B = 0 resulted in a reduction in power of the carrier. With this setup, for example, I could generate a zero bit as follows:

void gen_zero() {
    OCR1B = 0;
    _delay_ms(200);
    OCR1B = 166;
    _delay_ms(800);
}

After I had this set up, I implemented functionality to broadcast WWVB-format data by repeatedly broadcasting the appropriate data for the current second and then incrementing the current time.

I wanted to keep the physical design simple, so I opted to go with a press-fit design consisting of a 3D-printed top and bottom with laser-cut sides to form a box.

I used OpenSCAD, a programming-based 3D modeler, to design my 3D parts:

I used a Stratasys uPrint SE to print my parts out of ABS thermoplastic:

I used Adobe Illustrator to design my 2D parts, and I cut them out of acrylic on a 75-watt Universal PLS 6.75:

Because it was a press-fit design, assembly took about two minutes! Here’s the final product:

μWWVB works really well for me, consistently synchronizing my watch in about three minutes. My watch is set up to automatically receive the WWVB signal every night, so by leaving my watch on its stand overnight, it’s automatically synchronized every day!

In the current implementation, μWWVB syncs my watch to an accuracy of about 500 milliseconds of UTC. By putting a little more effort into making the timing in my software more precise, doing things like using the milliseconds value from the ZDA NMEA message instead of ignoring it, I could probably get the error down to about 100 milliseconds. There would still be some error, mostly due to the ZDA NMEA message being sent over UART, which is an asynchronous connection.

If I wanted the system to be much more accurate, I’d probably need to switch to a pulse per second (1PPS) GPS. A 1PPS GPS outputs a signal that has a sharp edge every second precisely at the start of the second — such a signal could be used to clock the WWVB time code such that each bit starts precisely at the start of the second.

But for now, for my purposes, μWWVB works really well!

You can find the board design, CAD designs, and source code for μWWVB on GitHub.