[time-nuts] Why are 1PPS signals so skinny?

Attila Kinali attila at kinali.ch
Tue May 15 22:40:07 UTC 2012


On Tue, 15 May 2012 16:51:13 -0400
Mike S <mikes at flatsurface.com> wrote:

> On 5/15/2012 4:19 PM, Attila Kinali wrote:
> > If the
> > PPS pulse is short, it contains very little energy, which means
> > the energy can be supplied by the small capacitors at the output
> > driver. The longer the pulse gets, the more energy it needs.
> 
> The pulse is meaningless. It's only the leading edge that matters. I 
> understand how shorter pulses may make for marginally cheaper electronics.

It's not marginally, it's quite substantial.
A 10nF capacitor can easily provide enough power for a 1-20us pulse
(rough guestimate). For a 1ms pulse, you need a 1000 times larger capacitor,
ie 10uF. The pulse rise times we want to acheive are <1ns, which means
we have frequency components far into the GHz range. It is easy to find an
10nF capacitor that is good to a couple of GHz. Good luck in finding
a 10uF that goes over 3MHz, and even these are not cheap at all.
Yes, you can combine a 10nF and a 10uF so that the 10nF provides the
energy for the edge and the 10uF the energy for the pulse, but this
again is more difficult to do correctly [1]. You also have more than
1000 times higher capacitive loading to your power supply, which
means you have to account for a roughly 1000 times higher rush in current,
which makes your power supply at least 10 times more expensive. You have
to choose a different driver that can handle the output power. And this
driver is either a lot more expensive or slower, most likely both [2].

Or in short: there are quite a few compromises in the design of such
a circuit. Each part has to fit to the others, or you will not acheive
optimal performance.

The big reason why we can have cheap rubidiums at home these days is not
because they are build in large quantities and thus have become cheaper.
No, it's because technological advances made it possible to do less
or simpler compromises in the design, which made the device cheaper,
and thus more devices got sold. Same goes for virtually any other
"high tech" device you have in your home. Starting from your fridge,
over your TV set up to your computer.


> > Which might have a negative effect on their performance.
> 
> I might win the lotto. The question is exactly _how_ does it effect 
> their performance, especially if they're synchronizing to the PPS signal.

We are measuring timing differences in the range of a couple of ps. This
is faster than the rising time of the pulse you have. Ie the steepnes
of the pulse is very important. Any modulation due to the power supply
being a percent higher or lower will be visible in the measurements.
And a percent variation is quite a good value for a "normal" power supply
design with large load steps.

Or think about your oscillator. Your power supply varies, this means
the power supply of the oscillator varies too. This will ever so slightly
(or not so slightly) modulate the drive strength of the oscillator which
in turn will shift the frequency. And i'm not yet talking about the modulation
that occurs on the EFC due to imperfect power supply rejection in the
circuit that produces and processes the EFC.
 
> > it's no use of having a fast
> > rising edge, if the pulse colapses a couple ns later.
> 
> Huh? If ns is too short, and ms is too long, what makes us just right? 
> And why are there so many timing receivers that only output on the order 
> of 20 us, when there are so many inputs which may require a few ms?

If your capacitor is too small, it will only be able to provide the
energy for the rising edge. After that, the capacitors further back
have to support it. If they have a too high impedance, the voltage at the
output driver will drop after the first edge. If the drop is too soon, the
receiver of the pulse will not detect the first rising edge, but trigger on
the second, washed out and slow edge.

There are draw backs even if there is no drop, but a limit in the height
the pulse reaches. If, due to too small capacitors, your pulse does not
reach the full 5V, you have to use a smaller trigger voltage for your
receiver. Which means you have a lower noise marign (the output and input
noise voltages are often independent of drive strength). This in turn
means you have more jitter on your pulse on the receiver side. Which
will limit the resolution you get for your instruments.

> 
> PPS is edge triggered, not level triggered. 

Yes, but the receiver has a minimum length requirement for the pulse,
otherwise it will not see it. Or even worse, it will not trigger correctly
and might enter a metastable state, which can lead to:
1) a delay of several ms (yes, miliseconds!) until the device 
correctly triggers,
2) high current consumption within the receiver due to high and low side
transistors conducting at the same time, which in turn will lead to
heating and might in the worst case destroy the receiver, in the best
case you just have degraded performance due to higher thermal noise.

> Once the leading edge is 
> transmitted (and it by necessity has a very fast rise time, so it looks 
> to capacitors, transformers, etc. as a high frequency signal), the shape 
> of the pulse really doesn't matter much.

It would not matter in an ideal world. Unfortunately, this world is not
ideal.

> Some devices need more than a 
> minimum above some threshold, but what ones need less than a maximum? If 
> it doesn't look like a flat topped pulse, so what? As long as the decay 
> is basically monotonic, and the receiver has some hysteresis (reasonable 
> assumptions), it makes no difference.

You have to fullfill the requirements of the receiver. Whatever those
might be. And they can be very difficult to reach at times.


Or to look at it from a different angel:
These days, we are spoiled. A lot! We buy on ebay stuff for "home" use
that engineers would have killed for just 10-15 years ago. And we get
it very cheap too! (Just think about the FE-5680 we all bought for
40USD). We are able to perform measurements at home that were very
challenging to do in a very well equiped lab 20-30 years ago. We have
reached an area of the time-nutty things we do, that even the slightest
pertupertions of the most unimportant signals can lead to degraded
measurements.

Just have a look at [3]. The spurs you see there are most likely
* 60Hz mains
* 120Hz mains (first harmonic)
* a "nearby" radio station (according to TVB)

And this in the measurement of an high performance oscillator with
very good power supply and well shielded cables.

[4] is also very interesting in terms of "how sensitive is that stuff?"


I hope this sheds more light on why things are not as easy as they
seem at first.

			Attila Kinali


[1] Your two capacitors form a nice L-C tank with a quite high Q, which you
drive with a nice high energy step function. This is because a capacitor
above its resonant frequency behaves like an L. Ie the small capcitor will
behave as a C and the bigger one as an L. As the equivalent series
resistance of capacitors with high frequency rating is very low, you get
a very high Q.

[2] The power rating for CMOS transistors is roughly dependend on the width
of the gate. The switching speed is roughly inverse proportional to the
capacitance of the gate (as long as the rise time of the gate drive is
slower than the transition time of the majority carriers, which it is in most
cases). The gate capacitance is in turn proportional to the area, which is
proportional to the width if the length is kept constant.
Or in short: a more powerfull transistor is slower. Or a faster switching
transistor cannot only drive a fraction of the current.

[3] http://febo.com/pages/oscillators/fts1050a/fts1050a-hp5065a-28day-spectrumplot.png

[4] http://www.febo.com/pages/oscillators/wenzel_uln/supply.html
-- 
Why does it take years to find the answers to
the questions one should have asked long ago?



More information about the time-nuts mailing list