Show us your square wave

[Jay_Diddy_B]

Hi,

Let me take to extremely bad cables. Lets us try 2m (6.5 ft) of lamp cord 2 x 18 awg:



At 10 MHz with 4.5ns rise and fall times you can see several reflections in the cable:



With the frequency changed to 2 MHz and the rise and fall time increased to 30ns the waveform doesn't look too bad:




I have posted these to reinforce the concept that the impedance is less important if the propagation time is less than the rise time.

Regards,

Jay_Diddy_B

[katzohki]

I wonder if we could glean any interesting information from this discussion in terms of the frequency domain...

I'm picking up an HP 8116A and HP 8904A on Friday so I'll try to post some pics coming Monday. Someone already posted with the 8116A, it looked pretty good I think.

[T3sl4co1l]

I wonder if we could glean any interesting information from this discussion in terms of the frequency domain...

I'm picking up an HP 8116A and HP 8904A on Friday so I'll try to post some pics coming Monday. Someone already posted with the 8116A, it looked pretty good I think.

Certainly can: the time and frequency domains are dual: complementary and inverse.  Shorter risetime means higher bandwidth; wave shape is determined by harmonics (fundamental frequency * N), while envelope shape (e.g., AM modulation) is determined by sidebands (fundamental frequency +/- X), the rate of which corresponds to the difference and so on.

Because of this dual property, one can use the time domain response to derive an equivalent RLC circuit (assuming certain conditions), just as one can use the frequency domain response.

Tim

[joeqsmith]

Rise time is measured at just over 5ns, which is pretty good. (Bandwidth can be approximated as 0.35 / (rise time), so we can't expect any faster than a 5ns rise on a 70MHz scope.)

I would have guessed that rule of thumb was flushed down the toilet long ago.

Is this true?

What do the specs show for DSOs made in the last 10 years?   

[Hugoneus]

I'll play. How is this?  O0

[rx8pilot]

54ps...nutz.

I think the 8Ghz instrument sees a lot.

[tautech]

54ps...nutz.

I think the 8Ghz instrument sees a lot.

That is on no way the impressive stat........2.5 GHz square wave.   

I think you win  Hugoneus

[rx8pilot]

That is on no way the impressive stat........2.5 GHz square wave.   

I guess that is reasonably quick.
Now I am curious what 6g/bps HD-SDI looks like at 100 meters. That would stress even that fancy Agisight scope.

What generated the 2.5 Ghz and how was that signal delivered to the instrument?

[miguelvp]

54ps...nutz.

I think the 8Ghz instrument sees a lot.

That is on no way the impressive stat........2.5 GHz square wave.   

I think you win  Hugoneus

I don't know.... I think I see some Jitter, maybe it has a non locking PLL


If the wink is not enough, I kid 

[Yago]

Hehe, Shahriar entered the thread like Clint Eastwood in a Fist Full of Dollars!

[T3sl4co1l]

Hehe, Shahriar entered the thread like Clint Eastwood in a Fist Full of Dollars!

Well, considering halfway up the thread, others have posted their 30ps waves from SD-21 or whatever they are TDR rigs...

Tim

[TunerSandwich]

I think some tests with controlled impedance lines would really be in order here.  I still think it's a bit of a pissing contest, or at least doesn't have much practical application.....but I can see the fun in it. 

I dug out some spools of NiCr, Kanthal A-1 and some of our own proprietary Ti resistance wire.....using these as a basis for conductors, might lend some insight into how VERY controlled impedance changes these measurements, as well as how the tempco changes measurements....especially as the transmission lines polarize and saturate.....

Maybe, just for fun, throwing some controlled capacitance into those transmission lines would also lend to some insights....

I think i have some PTFE jacketing laying around, as well as some shielding braid.....but that might start getting a bit carried away.....

I will build up some controlled impedance transmission lines....with no shielding, and limit the freq.....to hopefully show that conductor impedance mismatches skew these measurements more than just about anything....and RF inductance be damned 

I have personally never done that experiment, but I think theory is well established/sound....I am curios to see how sub ohm discrepancies between conductors skew the measurements.....especially with low energy signals....

any suggestions or ideas on what resistance offsets to attempt?  what amplitude signal?  timebase/freq?  etc....

I think I can build up some pretty "bad" cables here....and see just how far we can decimate a square wave.....

[T3sl4co1l]

Sure, that's easy enough to do...



The subject: 50' of lamp cord.  BNC to binding posts on either end.

Oh, and just for kicks, I connected one end backwards.



Didn't even bother with the fast risetime on this one; the bounce is obvious.  It's also unterminated.  Overshoot suggests Zo ~= 100 ohms, which is typical for twisted pair and the like.

On the faster rise time, I get 55ns rise terminated, 28ns unterminated.

Due to the twist, at lower frequencies, the square wave tilts down towards ground, as if shunted by an inductance, because... it is.

Another good exercise is to take a 10x probe near sub-10ns edges and see what probing does.  Without any ground clip at all.  You may be pleasantly surprised by the outcome!

Tim

[TunerSandwich]

Yes, but that is copper.....the NiCr or A=1 our our own fancy Ti resistance wire would be a lot of fun, because the polar offset can be controlled to around 0.1ohm per foot.....

I am curious to see if the scope can resolve a 0.5ohm or lower offset, through 50 ohms.....with let's say a 10mA 1K pulse....

Also can we rectify some of the sine waves that make up the pulse (with no diode ICs), and actually get a series of pulses within the pulse....

I am curious about what saturation level in the Ti wire (especially) will lead to some feedback, and introduce some controlled oscillations

Worlds crudest digital multiplexer? Or just a shitty theory? 

[Yago]

Hehe, Shahriar entered the thread like Clint Eastwood in a Fist Full of Dollars!

Well, considering halfway up the thread, others have posted their 30ps waves from SD-21 or whatever they are TDR rigs...

Tim

Yes, sorry it was my light hearted frivolity, and lack of understanding in this area.
Didn't mean a disservice to the thread.

[Hugoneus]

Hehe, Shahriar entered the thread like Clint Eastwood in a Fist Full of Dollars!

Well, considering halfway up the thread, others have posted their 30ps waves from SD-21 or whatever they are TDR rigs...
Tim

Yes, sorry it was my light hearted frivolity, and lack of understanding in this area.
Didn't mean a disservice to the thread.

Alright, alright... How about an 81Gb/s circuit I made a while back with a rise time of 5.2ps?  O0

Measured using an Agilent 70GHz remote heads on a sub-sampling scope, on wafer with 1mm probe and cables.

[G0HZU]

In theory at least, I think that the characteristic impedance of the transmission line becomes irrelevant at certain cable lengths if the squarewave is continuous. Because a square wave is made up of a series of sine waves of F, 3F, 5F, 7F etc etc then the characteristic impedance of the line doesn't matter if the line is (for example) 180degrees long at F.

So if you had a (continuous) square wave that had a very fast risetime then you can still get a good square wave if the line is 180 degrees at frequency F. So in theory you can break the risetime vs propagation time rule if the cable is a certain length (or multiples of this length).

In reality, a typical transmission line will not behave as ideally as this but at high clock frequencies it might be worth experimenting with the length of the cable  at the clock frequency F. A good length to try is 180degrees at the clock frequency.

I still think it's a bit of a pissing contest, or at least doesn't have much practical application.....

Agreed. It does all seem a bit silly. The fastest pulse generator I have here was made in the 1970s and cost me £5 and that has been good enough for me for the last 20years or so

[TunerSandwich]

In theory at least, I think that the characteristic impedance of the transmission line becomes irrelevant at certain cable lengths if the squarewave is continuous. Because a square wave is made up of a series of sine waves of F, 3F, 5F, 7F etc etc then the characteristic impedance of the line doesn't matter if the line is (for example) 180degrees long at F.

So if you had a (continuous) square wave that had a very fast risetime then you can still get a good square wave if the line is 180 degrees at frequency F. So in theory you can break the risetime vs propagation time rule if the cable is a certain length (or multiples of this length).

In reality, a typical transmission line will not behave as ideally as this but at high clock frequencies it might be worth experimenting with the length of the cable  at the clock frequency F. A good length to try is 180degrees at the clock frequency.

I still think it's a bit of a pissing contest, or at least doesn't have much practical application.....

Agreed. It does all seem a bit silly. The fastest pulse generator I have here was made in the 1970s and cost me £5 and that has been good enough for me for the last 20years or so

Yes indeed....I posted on page 11 (some screenshots) showing this exact concept.  You can clearly see that the 15MHz square is made of over 35 discreet tones between DC (ok ok not DC, but I can't be bothered to go back and look at that....let's call it 0.1HZ) and 100GHz.  The rolloff was sharp past that, but the signal tones went out past 800GHz.  I did this measurement @ 200GS/sec (R.I.S.) and sin x/x interpolation....so the details aren't wholly trustworthy.....but do follow theoretical observations. 

I suggested using resistance wire to build some cables, because then we can make some accute predictions and observations....that have a basis in a measured set of benchmarks.  I.E.  what it the transmission line impedance discrepancy (closed loop) was a precise 0.1 ohm per foot?  At some point I am not going to sit here and laser trim the cable ends, but at least that is a ballpark measurement, and we aren't guessing about the impedance discrepancies from +/-.....

Bigger question though....would the scope even be able to resolve that offset, through it's own 50ohm shunt? And even then I wonder how much impedance discrepancy the actual sig gen out and scope in has (closed loop).  If we knew that, then we could derive a theory about how to swing and shape our pulses....

I suggested resistance wire, because it has a saturation point and I believe it will act as a diode at a certain point (zener effect).  I just think it would be hilarious to build a cable that rectified some of the square wave (sine) tones......It would be fun to see a square within a square, and turtles all the way down to the step response of the scope.....

[photon]

Hehe, Shahriar entered the thread like Clint Eastwood in a Fist Full of Dollars!

Well, considering halfway up the thread, others have posted their 30ps waves from SD-21 or whatever they are TDR rigs...
Tim

Yes, sorry it was my light hearted frivolity, and lack of understanding in this area.
Didn't mean a disservice to the thread.

Alright, alright... How about an 81Gb/s circuit I made a while back with a rise time of 5.2ps?  O0

Measured using an Agilent 70GHz remote heads on a sub-sampling scope, on wafer with 1mm probe and cables.

Shahriar,
This is fast. Is it silicon or some exotic semiconductor? What geometry? The speed of this clock (81Hz or 12.35ps) is faster than any I am familiar with. Can you share some implementation details?

[Hugoneus]

Shahriar,
This is fast. Is it silicon or some exotic semiconductor? What geometry? The speed of this clock (81Hz or 12.35ps) is faster than any I am familiar with. Can you share some implementation details?

This is an 81Gb/s TIALA-Retime in 65nm GP CMOS process. It uses a full rate clock, so 81GHz clock. You can see it here:

http://ieeexplore.ieee.org/xpl/abstractAuthors.jsp?arnumber=4674506

Cheers,

[photon]

Shahriar,
This is fast. Is it silicon or some exotic semiconductor? What geometry? The speed of this clock (81Hz or 12.35ps) is faster than any I am familiar with. Can you share some implementation details?

This is an 81Gb/s TIALA-Retime in 65nm GP CMOS process. It uses a full rate clock, so 81GHz clock. You can see it here:

http://ieeexplore.ieee.org/xpl/abstractAuthors.jsp?arnumber=4674506

Cheers,

Thanks for sharing your knowledge, much appreciated.

[T3sl4co1l]

Yes, but that is copper.....the NiCr or A=1 our our own fancy Ti resistance wire would be a lot of fun, because the polar offset can be controlled to around 0.1ohm per foot.....

The wha..?

I am curious to see if the scope can resolve a 0.5ohm or lower offset, through 50 ohms.....with let's say a 10mA 1K pulse....

Well, 0.5 ohm will add 1% of droop, roughly speaking.  Whether high or low frequencies are attenuated more, I'm not sure offhand how that factors in.

Also can we rectify some of the sine waves that make up the pulse (with no diode ICs), and actually get a series of pulses within the pulse....

I am curious about what saturation level in the Ti wire (especially) will lead to some feedback, and introduce some controlled oscillations

Tim

[TunerSandwich]

Tim

The resistance wire will saturate, and act like a diode.....and rectify some of the sinewaves in the pulse (zener effect)....can't be bothered to do the math to see if that will be in the bandwidth limitations of the original signal. 

I have done a lot of experimentation with those scenarios (part of what we do in my business) and most resistance wire will saturate to the point of oscillating loud enough to hear it (and that is just with DC).  Although I am not keen to dump 60 amps into my scope 

However if it rings loud enough to hear it at 60 amps, then I am most certain it rings at inaudible frequencies, with significantly less power. 

I will build up some cables out of this stuff and we can see if any of that proves to be true, with small signals......

Think ribbon tweeter.....

[T3sl4co1l]

I... don't think you know the definitions of those words...

Saturation is a characteristic of ferromagnetic and ferroelectric materials.  Resistance wires are almost never ferromagnetic.

Saturation has no effect on DCR (or, extremely little).

Saturation does not "act like a diode".

The zener effect is extremely specific: the breakdown of a narrow semiconducting junction due to tunneling current.  It does not appear in the physics of ferroelectrics or ferromagnetics.

Saturation is not oscillation, and oscillation is not DC.  Whatever instability you're talking about, it isn't related to any of the words you're using...

Ribbon tweeters use very well understood principles of electromagnetic induction and do not involve concepts of saturation, rectification, zener breakdown, or oscillation (aside from the intentional influence of acoustic waves).  The magnetic field in the device may be designed with ferromagnetic components, but this is not necessary (an admittedly very inefficient ribbon mic could be constructed with electromagnets and no ferromagnetic components).

Tim

[joeqsmith]

Rise time is measured at just over 5ns, which is pretty good. (Bandwidth can be approximated as 0.35 / (rise time), so we can't expect any faster than a 5ns rise on a 70MHz scope.)

I would have guessed that rule of thumb was flushed down the toilet long ago.

Is this true?

What do the specs show for DSOs made in the last 10 years?   

From Tektronix

Question:
How is bandwidth related to rise time for oscilloscopes?
Answer:

Historically, oscilloscope frequency response tended to approximately follow the rule: Bandwidth x risetime = 0.35. This corresponds to a 1- or 2-pole filter roll-off in the frequency domain. Today, at the high end, most real-time digital oscilloscopes more closely follow this rule:

Bandwidth x rise time = 0.45.

This corresponds to a much steeper frequency roll-off above the specified bandwidth. The steeper roll-off is more desirable in digital oscilloscopes that oversample by 4x, 3x, or even less because it prevents aliasing by eliminating any signal above the Nyquist frequency (1/2 the sample rate – the minimum sample rate required for accurate signal representation).

From Teledyne

Risetime-Bandwidth Product (RT*BW)
Typically RT*BW = 0.40 to 0.45 for modern high bandwidth scopes
Historically analog scopes RT*BW = 0.35