Frequency reference clocks should ..

[BillyO]

.. be:

A) Sine wave
B) Square wave
C) It depends
D) Should not really matter

[colorado.rob]

A "square wave" is an imaginary construct. If you think you need a square wave, you first need to specify maximum rise time. It is a sum of some limited number of sine waves.

A square wave is the EE equivalent of the physicist's "perfectly spherical cow".

[nctnico]

Square wave is better due to faster edges which cause less jitter. OTOH this could lead to reflections which distort the signal in poorly terminated cables (which is the case in many setups I've seen). The industry standard is sine wave.

[TimFox]

Many pieces of test equipment send the external reference input through a bandpass filter before further processing.

[BillyO]

A "square wave" is an imaginary construct. If you think you need a square wave, you first need to specify maximum rise time. It is a sum of some limited number of sine waves.

A square wave is the EE equivalent of the physicist's "perfectly spherical cow".

Close, but you got it bit backwards.  Being a physicist myself I'd have to say the "ideal square wave" is more akin to the "ideal cow" (whatever that would be). and the "spherical cow" approximation is more akin to the achievable real-world approximation of an "ideal square wave" 

[BillyO]

Square wave is better due to faster edges which cause less jitter. OTOH this could lead to reflections which distort the signal in poorly terminated cables (which is the case in many setups I've seen). The industry standard is sine wave.

Many pieces of test equipment send the external reference input through a bandpass filter before further processing.

So, the reason I'm asking is that I want to create a frequency reference.  To me it seems a lot easier to get a square wave OCXO and make a distribution amplifier using dirt cheap 74AC14 inverters than it is trying to make a 10MHz sine wave distribution amplifier.  I guess I could have some of the outputs left as properly attenuated and terminated square waves and could have some filtered down to sine wave.

[bdunham7]

Frequency reference clocks should ...

...not look like this?   (HP 5316B reference out)

[colorado.rob]

Frequency reference clocks should ...

...not look like this?   (HP 5316B reference out)

Show it with 50Ohm termination enabled.

[pqass]

Frequency reference clocks should ...

look more like this...   
HP 5334B with Option 010 High Stability Time Base (Oven) into 50R (confirms what the tag says: 0.5Vrms)

[bdunham7]

Show it with 50Ohm termination enabled.

Better, still not great.  I wonder if something is being overdriven here, the output is over 2X the 50mV_p-p that is specified.

[tautech]

FYI
10 MHz Ref Out from a $20k VNA with no OXCO installed. (default internal clock)

[Bud]

If you think logically for a sec, you want to lock your receiving device to Fref. You do not want it to lock to 2Fref, 3Fref, etc. Therefore having a clean sine wave Fref makes sense.

[TimFox]

Some of my old equipment, internally using a 10 MHz reference, would allow an external signal at 5 or 10 MHz, since it was passed through a narrow-band (probably crystal) filter.
Also, 5 MHz high-stability oscillators were a thing back then.

[nctnico]

If you think logically for a sec, you want to lock your receiving device to Fref. You do not want it to lock to 2Fref, 3Fref, etc. Therefore having a clean sine wave Fref makes sense.

That is only a problem if you have lots of distortion due to impedance mismatch. First thing that is done to the incoming signal is to turn it into a square wave.

[eloso]

Better, still not great.  I wonder if something is being overdriven here, the output is over 2X the 50mVp-p that is specified.

It will look better if you use the 50ohm input of your scope (like you just did) but now  discard your 1M probe. Use a piece of 50ohm cable.   Make sure you don't exceed your scopes max input which will be only a few volts.

Or put a 50ohm external load on the ref signal and then use your 1M probe across the external 50ohm terminator.

Interesting thread !

Eloso

[edpalmer42]

If you think logically for a sec, you want to lock your receiving device to Fref. You do not want it to lock to 2Fref, 3Fref, etc. Therefore having a clean sine wave Fref makes sense.

It depends what you want to use the reference for.

If all you're interested in is frequency, many pieces of equipment used some variation of a comparator to square up the reference so only the zero crossings were significant and a 'distorted' sine wave or square wave might be desirable.  But if you wanted to multiply the reference, then you might want a clean sine wave.

A common architecture has the internal reference phase-lock to the external signal.  That gives you a known quality reference - the internal one - that matches the frequency of the external reference.

Some equipment will lock to harmonics or subharmonics, some ignore them, some have that as an option.  Input impedance for the Ref-In jack varies from 50 ohms to hundreds of Kohms.  Amplitudes are similarly variable.  I typically run a daisy-chain from unit to unit and then terminate the chain with a 50 ohm terminator.  (Anyone remember 10Base2 ethernet?) 

Ed

[BillyO]

If you think logically for a sec, you want to lock your receiving device to Fref. You do not want it to lock to 2Fref, 3Fref, etc. Therefore having a clean sine wave Fref makes sense.

A square wave is a superposition of the fundamental and it's odd harmonics at specific amplitudes.  When we say it's a superposition it means their amplitudes are added together.  Those frequencies are not there on there own, so unless there is some strange filtering or math (like a Fourier transform) going on those harmonics will not be apparent in the waveform.  Logically then, it does not follow that it has to be a sine wave for the reasons you give.  And if there was filtering going on, it would be fairly steep filtering at precisely the required frequency.  Jus' sayin'

[TimFox]

As I mentioned above, if the external reference input (driven by your reference clock) goes through a band-pass filter, which would be fixed-tuned, it only has to be good enough to be far down at twice the reference frequency, and only at the third harmonic for a good (symmetric) square wave.

[BillyO]

As I mentioned above, if the external reference input (driven by your reference clock) goes through a band-pass filter, which would be fixed-tuned, it only has to be good enough to be far down at twice the reference frequency, and only at the third harmonic for a good (symmetric) square wave.

So 18dB/octave .. 24dB/octave?  This would only really be needed if you were expecting substantially less than ideal signals.  A well behaved Sq.W. should have a nice clean rising edge to trigger the phase measurement off as would a pure sine/triangle/saw.  But real life being real life I suppose these inputs need to be designed to extract the fundamental from potentially sub-par signals.

I've looked at the specs of the equipment I have to see what they demand for an external reference signal and all they ever give is a frequency and a minimum signal level.

Below is the signal I get from my HP 5300B.  It's not symmetrical, but nothing I have connected it to yet has complained.

[Detlev]

Hi all,

Last year I tried to determine the jitter of audio DACs. At the time, I noticed that square-wave signals have a significantly better Allan-deviation than sine signals.

During the setup, the SDG6022 and the FCA3000 are both attached to the Leo Bodnar GPSDO. Nevertheless, the time deviations in the signal for sine waves are worse by a factor of almost three. The Allen deviation is 150ps for sine and 60ps for square.

I think the voltage noise is noticeable due to the flatter edge and worsens the triggering of the connected device. So I would always prefer a square to a sine.



Have a nice weekend 
Detlev

[nctnico]

Hi all,

Last year I tried to determine the jitter of audio DACs. At the time, I noticed that square-wave signals have a significantly better Allan-deviation than sine signals.

During the setup, the SDG6022 and the FCA3000 are both attached to the Leo Bodnar GPSDO. Nevertheless, the time deviations in the signal for sine waves are worse by a factor of almost three. The Allen deviation is 150ps for sine and 60ps for square.

You have to be carefull with what you are measuring here. The noise of the input of the counter or the actual signal? IIRC your counter has a 100ps resolution only so you are measuring well into the noise floor. For measurements at these low jitter levels, a DMTD is a better tool.