Fast edge oscillator - traces length

[Chupacabras]

Hi,
I am building "Fast edge oscillator" that is described in this video:


I designed it in Kicad.
Does length of tracks matter? There are multiple tracks that are "joined" at the end and so create output to BNC.

[Redcat]

In simple words...Yes, lenght matters  .

Outputs are joined to get (roughly) 50 Ohm output impedance and match the cables impedance.

You will get delays and your signal will be not that clean as when you bodge the resistors with matched lead lenght.
So actually your etched ciruit may perform worse than this solution.
Have you ever seen tracks on a pcb snaking around? That's to match the lenght of the tracks because of delays. Of course you could do so too.

Very nice design btw you did. So try to match the lenghts and it will be perfect.
(It will work anyway ok).

Try to play with the cap and resistor value for the oscillator. The frequency is like 3 to 6kHz (when i remember correct) and doesn't matter too much, but you want the fastest edges possible.
(I have build this thing too, long ago...the bodged version  ).

Search the other video of Alan. I remember he has compared this (his) circuit to another one which was made clean and nice but didn't perform as well as his.

In the series of 74 hex inverters there are slower and faster chips availiable. Pick a fast series like the 74hc14 or a faster and rise time should be in the area of 3-7 ns. I see, you did in the schematic, thats good.

There are even much faster edge generators which work with an avalanche voltage on a transistor. Pulses with sub-nanosecond rise and fall times.
(Like http://www.aholme.co.uk/Avalanche/Avalanche.htm for example...just google" Avalanche Pulse Generator".)

Just watch Alans videos and you will find an example for this too (btw i highly recommend his channel).

Have fun, Tom



 

[Zero999]

Are the resistors really necessary?

The drawing says 74HC which has an output impedance of around 50Ohm so just use one of them as the output buffer and not bother with the resistors.

If you're using the 74AC series which apart from being faster, has a lower output impedance so the resistors are a good idea.

[Chupacabras]

Thanks Tom for reply. And thanks for compliment )
Actually I am a beginner, so it is good to know that I am going the right way

I am about to use 74AC14 which is even faster than 74HC14. I used HC version in Kicad because it does not contain AC one. (not so important)

I know that lengths matter in high frequencies. But I was not sure if it is necessary in this circuit.
I don't know how much will this improvement pay off in this circuit.

Thanks for the tip. I will definitely look at that avalanche pulse generator.
What channel do you mean? Some on youtube? (I got it, Alan=w2aew, I already am subscribed )

[bktemp]

Paralleling CMOS gates for highspeed outputs is a bad idea. At least it made everything worse for the circuits I have used:
I needed to buffer a clock signal around 100-150MHz to drive a 50ohm load. To do this I used 3 gates from a SN74ALVC244 buffer in parallel each with a series resistor to get a total 50ohm impedancel. All traces were matched to <1mm difference. It worked fine up to around 60MHz then it started to behave weird. The jitter was extremely high and the output frequency even startet to skip pulses at some frequencies. I ended with only using a single gate and this worked fine up to 200MHz.
There is another trap when driving loads using CMOS buffers: Ground bounce. Since Vcc and GND are on opposite side of the package, they have quite a large inductance. This may be the problem why paralleling gates does not work: As soon as one output switches, the ground bounce adds an offset to the whole ic and therefore effects the other channels.
I have seen circuits using the outputs as additional GND or Vcc pins by connecting them to GND or Vcc and driving the inputs to the appropriate level. The fets may have a higher dc resistance, but they are close to other ouputs and have therefore a much lower inductance.

[tggzzz]

To get a fast edge, I used 74lvc1g14 devices in a similar configuration, three in parallel giving a roughly-50ohm output, and 3 in parallel driving nothing at all.

Now I've only got a 100MHz scope - which is grossly insufficient to measure the edges. Not being made of money, I had to use some ingenuity (and £15/$20or so) to measure the rise/fall times, and they appear to be around 625ps.

See how I did it at https://entertaininghacks.wordpress.com/2015/08/11/measuring-digital-signal-edge-rates-without-an-oscilloscope/

[T3sl4co1l]

Doesn't matter in the least.

Your trace mismatch will be something like 100ps, and the gate match is several ns.

Tim

[Redcat]

Ah ok, nice information T3sl4co1l. I didn't know exactly how much. But i've really seen it perform "worse" than flying construction.
It will ring a little anyway.

Yes, the device will work as i said with non matched lenghts okay too ... i thought Chupacabras could make it perfect  .

There was a reason (apart from the 50 Ohm) which i can't remember  why using more than one inverter as buffer was said better....

And i'm pretty sure, I have used the 74AC.. too  .

When you finish your device and please show some photos. I bet it will look very nice. 

[Chupacabras]

Well, I made all the parallel tracks equal in length.
+/- 0.1mm
Preview is attached.

I will definitely make Avalanche pulse generator too.
I think it is a good way for me to learn

[Chupacabras]

There was a reason (apart from the 50 Ohm) which i can't remember  why using more than one inverter as buffer was said better....

I was curious about that too.
I think it is because single output of schmitt trigger is able to drive 75mA. But if there is 5V power supply and 50ohm impedance, it is 100mA, right? So at least couple of them are needed. That is my theory, someone could confirm that.

[tautech]

There was a reason (apart from the 50 Ohm) which i can't remember  why using more than one inverter as buffer was said better....

I was curious about that too.
I think it is because single output of schmitt trigger is able to drive 75mA. But if there is 5V power supply and 50ohm impedance, it is 100mA, right? So at least couple of them are needed. That is my theory, someone could confirm that.

There are many threads on " Avalanche pulse generator", search for them.
One from a couple of years back:
https://www.eevblog.com/forum/projects/transmission-line-avalanche-pulse-generator/

[T3sl4co1l]

Personally I might prefer AC04 to AC14: the Schmitt trigger inputs have slightly different input thresholds, therefore, slightly different timings.  This is still true of any gate, but might be important enough.  The actual propagation delays should be fairly consistent through the rest of the device.

Remember the HC family and relatives are input and output buffered CMOS types, so an inverter internally consists of three gates chained; a Schmitt trigger has an additional inverter and a large resistor to provide positive feedback to the output of the first inverter.

It would be worthwhile squaring up the input signal first, perhaps with a single 74LVC1G14 or something like that.  Then followed with a few regular inverters (AC or LVC family) in parallel, their outputs again connected with resistors.

The output resistance of a bare AC or LVC gate (at >= 3V) is typically 20-40 ohms.  The low resistance suggests it should be quite easy to connect to a transmission line, but a gate is a fairly small output, not intended for continuous half-shorted operation, and the variance in resistance means poor matching unless you swamp it with a more accurate resistor.

So what's normally done (and which is the purpose of the above circuit), is to add a series resistor, to reduce gate dissipation and get the tolerance useful (a +/-10 ohm variance means it's within 10% for a total resistance of 100 ohms; 5% for 200; or a series resistor of 70 or 170 ohms).  But this resistance is now too high to match, so several channels of this are connected in parallel (e.g., 4 x 200 ohms).

Tim

[Chupacabras]

I simply had to build 2 versions of this oscillator. One with unequal traces (attached pcb design in my first post) and the other one with traces equal in length.
The rise time is ABSOLUTELY the same. No difference whatsoever. Both has 3.2ns rise time.
And ringing looks the same in both.

Just a little difference in oscillation frequency, probably because capacitors are not perfectly equal. One has oscillation frequency 3.19Mhz and the other one 3.09MHz. But that is not important at all.

[tggzzz]

I simply had to build 2 versions of this oscillator. One with unequal traces (attached pcb design in my first post) and the other one with traces equal in length.
The rise time is ABSOLUTELY the same. No difference whatsoever. Both has 3.2ns rise time.
And ringing looks the same in both.

Just a little difference in oscillation frequency, probably because capacitors are not perfectly equal. One has oscillation frequency 3.19Mhz and the other one 3.09MHz. But that is not important at all.

Actually, you don't know the rise time. The rise time of your 100MHz scope is ~3.5ns. Therefore the signal rise time is significantly less.

[Chupacabras]

You are right. I have 100MHz scope.
So I should test it with some 200Mhz (and more) scope?

I am about to build Avalanche Pulse Generator.
So my scope will show 3.2nS 3.2ns (thanks tggzzz for correction ) rise time for it too?

[tggzzz]

You are right. I have 100MHz scope.
So I should test it with some 200Mhz (and more) scope?

There are other ways, as I indicated in my earlier post.

I am about to build Avalanche Pulse Generator.
So my scope will show 3.2nS rise time for it too?

I would hope it would show 3.2ns, since nS is a measure of conductance But yes, it would show 3.2ns.

You would benefit from reading the refs in https://entertaininghacks.wordpress.com/library-2/scope-probe-reference-material/ particularly AN-47 p21