Power draw of the clock, and getting it as low as possible while maintaining good accuracy and reliable operation with varying line voltage, were major design goals of the project. The first 2 tables below are related to power consumption under different operating conditions, and the next ones show how much the regulated supplies vary with line voltage. The shunt regulator tubes are somewhat starved for current at low line, which isn't the best for regulation, and increasing the current isn't good for power draw or regulator life, but that's one of the compromises I had to make.

When the clock was finished I also wanted to try some some 1960s solid state 6X4 replacement rectifiers to see how much lower the power would be without having to heat rectifier filaments. The downside of the solid state diodes is, of course, is that they might blow out in an EMP event. Oh well, if that happens at least it's easy to just pop the 6X4s back in.

Power consumption vs line voltage and rectifier type. Crystal oscillator and oven power on, clock running on internal reference. Power in watts (W):

Line voltage

110V

120V

130V

6X4 rectifier tubes

30

39

47

Solid state replacement rectifiers

26

33

39


Power consumption vs line voltage and rectifier type. Crystal oscillator and oven power both off, clock running with line frequency reference. Power in watts (W):

Line voltage

110V

120V

130V

6X4 rectifier tubes

23

30

39

Solid state replacement rectifiers

19

25

33


Power supply voltage variation vs line voltage. 6X4 rectifier tubes:

Line voltage

110V

120V

130V

+450V

439

447

449

+300V

294

297

299

+200V

197

199

201

+150V

148

148

150

+55V

54.1

55.5

55.9


Power supply voltage variation vs line voltage. Solid state replacement rectifiers:

Line voltage

110V

120V

130V

+450V

445

448

450

+300V

299

298

300

+200V

200

200

201

+150V

148

149

150

+55V

55.5

55.5

56.0


Timing Performance, Crystal Oscillator Reference

The next 2 plots show the timing performance over a one month period, followed by 9 graphs of what various internal and ambient temperatures, and internal power supplies and line voltage were doing during that time. Note that the graph labeled "Oven temp" is the outside of the oven assembly, not inside the temperature controlled area where the crystal is. Phase is the clock time. It varies between 0ms and -220ms over the 30-day run. The title of the frequency graph also shows the frequency drift per day, which is important to know if you want to adjust the clock to have as small an error as possible over a long period.

The influences of voltage and temperature are evident in the graphs, but it turns out that the voltage and temperature effects combine to affect frequency in a fairly complex way. Temperature is dependent upon voltage because higher voltage = more power dissipation = higher temperature. Also, the regulated voltage out of the gas tube regulators varies with temperature and over time. For example, a step change in the AC input doesn't produce a simple step change in the regulated voltage. There is an initial step, and then a slope in voltage over maybe a few hours as the tube settles. Just establishing a figure of merit for the gas regulators, like line regulation, isn't as simple as I thought it would be. I'll continue collecting data and trying to improve the oscillator performance if I can.



Timing Performance, Power Line Reference

When using the 60Hz line as a reference, the clock doesn't add any significant instabilities of its own due to voltage, temperature, or anything else to the time it keeps. Any other clock running on the same power grid, Western US in this case, will keep the same time. How good is that? The next plot shows the phase variations during the first 11 days in July, 2014 to be about 14s peak-peak. I've been collecting power line phase data like this continuously for over a year, and I can say that the variations in this plot are typical of what I've seen during that time. So if you set your power line referenced clock at just the wrong time, you could easily see an error of 14s, or more, in a few days.


TDEV is a measure of how much a clock's time can be expected to wander around over a given observation interval. The plot below compares the stability of the clock's oscillator vs the power line. It shows that if you watch both of them over a period of a day, 86400 seconds, the power line is stable to about 1s, and the crystal oscillator is stable to about 2ms, 500 times better. TDEV is a statistical measure for making comparisons, kind of like standard deviation, so this doesn't mean that the maximum error of the clock in one day is 1s for the line and 2ms for the crystal. All that can be said is that for the data set I show here, the clock running on the line frequency varied 14s peak-peak, and the crystal wandered 16ms peak-peak.


Next is the same data presented in terms of phase. Compared to the power line phase, the crystal oscillator time variation is so small it looks like a flat line right at zero.



I may be the only one, but I find the next 2 plots interesting becuase they show how much noise is added by neon (or any ionization based) circuits. These ADEV and MDEV graphs compare the frequency stability of the 1 PPS with the stability of the 36kHz crystal oscillator and the 6KHz analog frequency divider output. You can see that the 1 PPS, which has passed through an argon dekatron, 2 neon dekatrons, and 3 trigger tubes, is much noisier than either the oscillator or the analog frequency divider. This is because the timing of neon circuits depends upon the ionization time of the gas, which has a large random component. The ionization time and its stability can be lowered by adding just about any kind of energy to the gas. For example, the noise drops quite a bit when I run the clock outside the cabinet in a room with ambient light. Even with the window blinds closed, it's easy to see the effects of sunrise and sunset on the 1 PPS timing and noise. The trigger tubes are especially sensitive to the light from a blue laser pointer, and I can trigger the alarm from almost any distance as long as I can see V91, the alarm tube, and hit it with the laser spot.



Measurements and Data Collection

The clock timing measurements on this page were made by comparing the 1 PPS output of the clock to a local 10MHz reference steered long-term by GPS. Many hobbyists build their own 10MHz reference or use a commercially made GPSDO module, the Trimble Thunderbolt being a common one. I have several types of commercial test equipment for making the timing measurements, but for testing this clock or the power line frequency I don't need or want to use a rackmount counter with nanosecond or less resolution. A simple, low power solution is to use a time stamping counter like the PicPET. To make the PicPET more useful in my lab environment, I build it into a circuit using a Lantronix XPort. to give it network connectivity.

The temperature and voltage data was collected with a home made data acquisition system I made eons ago. I don't have a schematic, but I know it's a simple circuit using a 68HC05 micro and a TLC2543 12 bit, 11 channel serial ADC.

The last plots on this page were made with TimeLab, an indispensable software tool for time and frequency measurement and analysis.