Fast square wave generator-- bad ringing...

[George R. Gonzalez]

Sooo I'd like to build a fast square wave generator to test out some
scopes.

I wire up a 74AC04, two inverters and a R/C are the oscillator, runs at
around 10MHz,
another inverter as a buffer, then three inverters in parallel as the
output stage.

Output goes to a 10-1 divider to match 50 ohms (and coincidentally to get
the voltage
down to the level the Tek smapling plug-in can handle)


All with very short leads, several bypass capacitors, ranging from 470pf to
20,000 pf.

It runs fine, putting out square waves with really fast rise and fall times,
faster than
my old Tek <400ps sampling plug-in can resolve. (75 ps plug-in on order).

PROBLEM IS, there's TONS of ringing on what should be the flat tops!

Adding bypasses helps SOME, but it mainly just changes the character of the
ringing,
doesnt reduce the amplitude by much.

Is this one of those counter-intuitive situations where "hard" bypassing
isnt what's needed,
I have to add a few ohms of series resistance on the bypasses? Weird
concept for sure.
How many ohms? Or should I just find some very small 50 ohm pots to put in
series with the various size bypasses and just tune everything for best
flatness? What a weird evironment.


Any hints appreciated,

George

 

[A E]

Sooo I'd like to build a fast square wave generator to test out some
scopes.

I wire up a 74AC04, two inverters and a R/C are the oscillator, runs at
around 10MHz,
another inverter as a buffer, then three inverters in parallel as the
output stage.

Output goes to a 10-1 divider to match 50 ohms (and coincidentally to get
the voltage
down to the level the Tek smapling plug-in can handle)

All with very short leads, several bypass capacitors, ranging from 470pf to
20,000 pf.

It runs fine, putting out square waves with really fast rise and fall times,
faster than
my old Tek <400ps sampling plug-in can resolve. (75 ps plug-in on order).

PROBLEM IS, there's TONS of ringing on what should be the flat tops!

Adding bypasses helps SOME, but it mainly just changes the character of the
ringing,
doesnt reduce the amplitude by much.

Is this one of those counter-intuitive situations where "hard" bypassing
isnt what's needed,
I have to add a few ohms of series resistance on the bypasses? Weird
concept for sure.
How many ohms? Or should I just find some very small 50 ohm pots to put in
series with the various size bypasses and just tune everything for best
flatness? What a weird evironment.

Any hints appreciated,

George

I use a PC clock synthesizer in a box to get fast edges. But then, I use a 1S1
sampler, not anywhere close to a 75ps rise time. It would help to have a
schematic and pictures of your setup, since it's as much the construction as the
circuit that can be a problem.
http://www.dfpresource.org/winbondclock.html
The advantage of using a PC motherboard for your testing purposes is that the
nasty work of layout and bypassing are all done, and presumably tested. I like
the newest synths like in the AMD mobo I'm carving up now, they have
differential outputs.
How did you achieve the 10:1 attenuation? I like using pads, or MiniCircuits
attenuators, they are rated and specified to run at >1Ghz.
Anyways, I'd like to hear and see back from you, this stuff is fun.

 

[Winfield Hill]

George R. Gonzalez wrote...

Sooo I'd like to build a fast square wave generator to test out some
scopes. I wire up a 74AC04, two inverters and a R/C are the oscillator,
runs at around 10MHz, another inverter as a buffer, then three inverters
in parallel as the output stage. Output goes to a 10-1 divider to match
50 ohms ...
PROBLEM IS, there's TONS of ringing on what should be the flat tops!

Careful, what you want is a good 50-ohm source for your 50-ohm coax
to the scope right? So you want a 43 to 47-ohm series resistor in
series with the paralleled inverters. Use very short wiring to the
resistor and coax jack and keep the wiring and resistors very close
to your ground plane. Should be fine. If you want a smaller signal,
use coaxial wideband 50-ohm attentuators on the output.

You could try 250 ohms from the CMOS to 62.5 ohms to ground, for a
50-ohm 1V source for your BNC jack. With a 50-ohm termination at
the far end you'll get a 500mV pulse. Add a second stage if you need
more less than 500mV, don't try for too much attenuation in one stage.
BUT, unless you have small SMT resistors, etc., and are very careful,
you'll no doubt end up with poor frequency response and ringing. Good
coaxial attenuators are made using controlled impedances, and with
symmetrical rod and disc resistors, etc.

Thanks,
- Win

 

[John Popelish]

Sooo I'd like to build a fast square wave generator to test out some
scopes.

I wire up a 74AC04, two inverters and a R/C are the oscillator, runs at
around 10MHz,
another inverter as a buffer, then three inverters in parallel as the
output stage.

Output goes to a 10-1 divider to match 50 ohms (and coincidentally to get
the voltage
down to the level the Tek smapling plug-in can handle)

All with very short leads, several bypass capacitors, ranging from 470pf to
20,000 pf.

It runs fine, putting out square waves with really fast rise and fall times,
faster than
my old Tek <400ps sampling plug-in can resolve. (75 ps plug-in on order).

PROBLEM IS, there's TONS of ringing on what should be the flat tops!

Adding bypasses helps SOME, but it mainly just changes the character of the
ringing,
doesnt reduce the amplitude by much.

Is this one of those counter-intuitive situations where "hard" bypassing
isnt what's needed,
I have to add a few ohms of series resistance on the bypasses? Weird
concept for sure.
How many ohms? Or should I just find some very small 50 ohm pots to put in
series with the various size bypasses and just tune everything for best
flatness? What a weird evironment.

Any hints appreciated,

George

I often deaden a bypass capacitor by putting an ohm or two in series,
especially in series with a big cap in parallel with a smaller one.

But I think your problems may be aggravated by a poor layout. Can you
post a picture of your circuit and a schematic to
alt.binaries,schematics.electronic?

The bypass caps and the grounded resistor of the voltage divider have
to connect very close to the negative supply pin of the output
inverters. I would put the bypass cap under or over the inverter
pack, rather than around it.

 

[James Meyer]

PROBLEM IS, there's TONS of ringing on what should be the flat tops!

Adding bypasses helps SOME, but it mainly just changes the character of the
ringing,
doesnt reduce the amplitude by much.

You aren't by any chance using an "ordinary" probe with "ground" clip to
connect your generator to the scope, are you? If so, that's your problem. You
must use a 100% coaxial connection all the way from your generator to the scope.

Jim

 

[John Larkin]

Sooo I'd like to build a fast square wave generator to test out some
scopes.

I wire up a 74AC04, two inverters and a R/C are the oscillator, runs at
around 10MHz,
another inverter as a buffer, then three inverters in parallel as the
output stage.

Output goes to a 10-1 divider to match 50 ohms (and coincidentally to get
the voltage
down to the level the Tek smapling plug-in can handle)


All with very short leads, several bypass capacitors, ranging from 470pf to
20,000 pf.

It runs fine, putting out square waves with really fast rise and fall times,
faster than
my old Tek <400ps sampling plug-in can resolve. (75 ps plug-in on order).

PROBLEM IS, there's TONS of ringing on what should be the flat tops!

Adding bypasses helps SOME, but it mainly just changes the character of the
ringing,
doesnt reduce the amplitude by much.

Is this one of those counter-intuitive situations where "hard" bypassing
isnt what's needed,
I have to add a few ohms of series resistance on the bypasses? Weird
concept for sure.
How many ohms? Or should I just find some very small 50 ohm pots to put in
series with the various size bypasses and just tune everything for best
flatness? What a weird evironment.


Any hints appreciated,

George

Suggestions:

Build it flat on a small sheet of solid copperclad PCB material. Bend
the IC leads out straight. Solder the IC ground pin directly to the
plane, and use a 0.1 uF or thereabouts chip bypass cap directly from
Vcc to the plane.

Keep signal loop areas minimum. Think zero length.

Use chip resistors or 1/4w carbon comps in your output divider. Keep
them short and low.

Use a separate IC as the final output stage, and only use one section.
Pull up unused inputs and ground unused outputs; this will reduce
ground bounce. Your oscillator sections are now probably bouncing your
output, staggered in time.

Use a Fairchild UHS-series tinylogic buffer as the output stage...
much faster than AC. Hard to solder, though. Hard to see!

Use a crystal oscillator, so it can be a timebase reference, too.

A very clean, sub-ns step should be possible.

John

 

[Fred Bloggs]

What are the two inverters in series intended for?
--

He's talking about this EMI generator:

Please view in a fixed-width font such as Courier.




/\ /\ /\ /\
o o / \/ \/ \/ \ o o

VDD |
--- |
| |
| |
| |
| \ +-----------+
| \ |
+----| >O--+------+ |
| | / | | |
| | / | / |
OSC | | 450 |
| | / |
| \ | \ | \ | | \ | \ +->
| \ | \ | \ | | \ | |
+---| >O----+-----| >O--+--| >O-+----| >O--+ +->
| | / | | / | | / | | / | |
| | / / | / | | / | | / | /
| R | | | 51
| / | | | /
| \ | | | \ | \
| 5R | C | | | \ | |
+----/\/\-----+-----||------+ +----| >O--+ |
| | / |
\|/ | / -----------+
- 1.5VDD - -+-
| \ | \
| \ | \
| \ | \
| \ | \
\ | \ | \
\ | \ | \
| / | 0.5VDD/ |
| / | / |
| / | / |
-+---/----------|----/----------|---
| / | / | /
| / | / | /
|/ -0.5VDD | / |/


F~2.2RC

 

[A E]

Since you recommend a seperate IC package for the output buffer, what if
you want to put a solid 2.5V into 50 ohms, which necessitates using
several inverters in parallel (if you don't want to exceed 50mA per gate
if it gets shorted)? Do you think paralelling the gates causes so much
ground bounce from the summed switching currents that it is effectively
impossible to get really clean edges? Also what do you think about my
comment in my response to OP about the non-linearity of the gate output
impedance? I think that for unterminated lines in which the driver

Now *that* is an interesting subject. Output impedance is usually poorly
specified for logic chips at best. I find the specs of clock synth chips to be
much clearer in that respect. Usually, putting 50 ohms in series is optimistic,
the output impedance of a logic gate is not 0ohms, it's more like 10-40 ohms,
and that's non-linear on top of it all. That's why I also like putting 3dB pads
on the ouput of logic chips that are meant to drive 'pure' loads as opposed to
'impure' logic inputs. The pad tends to absorb reflections at the source and
presents a linear source impedance to the load.

output Z must eat the back wave, this is a problem, but reduces in
inverse proportion to how many gates are paralleled. But adding more
gates makes the ground bounce worse!

I think the outputs can't transition at the same exact instant, there has to be
some sort of short circuit current there... I'm gonna try to simulate that.

I posted a response to the OP and linked to some of my photos. You
might want to see what my waves look like on a breadboard. Note this is
by no means optimized from an *analog* perspective, only optimized
enough to give a clean edge for driving other digital gates.

But I am curious to learn more about generating baby-butt smooth edges,
just for interest.

My odd experience is that when I put a 74ACTQ14 in SO14 package on a
4-layer PCB with +5V and ground in the solid inner layers, with 0.1uF
0805 ceramic bypass cap right under the IC, that this setup gave uglier
edges than my breadboard!

I still have to figure out what went wrong with that.

Check the specs on that 0805 cap, and how did you get the signals out to the
scope/whatever?

Actually, I'm considering giving up on HC and AC digital single-sided 50
ohm line drivers. I am happier with the TC4427A and TC4426A MOSFET
drivers. A little slower, but therefore a little softer. But they are
also 4 times more expensive :-(

Hm, how many copies of the signal do you need? If you want some serious drive
and control, consider a PC clock driver, such as the Cypress W255H. It's free if
you can find the right PC mobo to disassemble.

 

[Dana Raymond, a minor God]

Without reading all of the replies to this post, I'd like to suggest that
you use a proper 50 ohm driver for your design. There are many 50 ohm
drivers available. Then make sure that your PCB layout also uses 50R rules,
that your connector is 50R and you may or may not need a 50R series
termination resistor.

Dana Raymond

 

[Paul Burridge]

He's talking about this EMI generator:

What's an "EMI generator"? I can only think of 'electromagnetic
interference' but can't imagine why anyone would want such a device.

 

[John Larkin]

Hey John:

Man, do you think like an RF designer or what!

Doing something at, say, 3 GHz is easy; you just resonate out the
parasitics. Wideband, DC-3GHz, is more inreresting.

Since you recommend a seperate IC package for the output buffer, what if
you want to put a solid 2.5V into 50 ohms, which necessitates using
several inverters in parallel (if you don't want to exceed 50mA per gate
if it gets shorted)? Do you think paralelling the gates causes so much
ground bounce from the summed switching currents that it is effectively
impossible to get really clean edges?

Every time a CMOS stage switches, there's a zot of shoot-through
current. That's why I think the oscillator stages should be separate
from the output, so the oscillator glitches (sooner in time than the
output) don't share the ground pin inductance. Similarly, the less
stages that switch, the better.

You can always power the chip from a current-limited supply if the
short-circuit current would be scairy.

Also what do you think about my
comment in my response to OP about the non-linearity of the gate output
impedance?

AC-class gates are pretty ohmic as long as they're within, say, a volt
of either rail. One or two sections are plenty to drive a 100 ohm or
higher load.

I think that for unterminated lines in which the driver
output Z must eat the back wave, this is a problem, but reduces in
inverse proportion to how many gates are paralleled. But adding more
gates makes the ground bounce worse!

I posted a response to the OP and linked to some of my photos. You
might want to see what my waves look like on a breadboard. Note this is
by no means optimized from an *analog* perspective, only optimized
enough to give a clean edge for driving other digital gates.

But I am curious to learn more about generating baby-butt smooth edges,
just for interest.

My odd experience is that when I put a 74ACTQ14 in SO14 package on a
4-layer PCB with +5V and ground in the solid inner layers, with 0.1uF
0805 ceramic bypass cap right under the IC, that this setup gave uglier
edges than my breadboard!
Mysterious!


I still have to figure out what went wrong with that.

Actually, I'm considering giving up on HC and AC digital single-sided 50
ohm line drivers. I am happier with the TC4427A and TC4426A MOSFET
drivers. A little slower, but therefore a little softer. But they are
also 4 times more expensive :-(

Take a look at Fairchild and OnSemi single-gate parts. Many are
sub-ns. If you power an NC7NZ34 from +6 volts, and parallel all three
sections, it will put over 5 volts into 50 ohms in well under 1 ns,
and make a very pretty edge. And they're dirt cheap.


John

 

[Jon Elson]

Output goes to a 10-1 divider to match 50 ohms (and coincidentally to get
the voltage
down to the level the Tek smapling plug-in can handle)

Are you sure all these resistors are non-inductive. Resistors around
50 Ohms often start to have spiral grooves in the element, and at
500 Ohms, they definitely have several turns of spiral. You may be
able to look at them under a microscope and see the profile of the
groove in the paint layer. You might get better results using chip
resistors (which at least can't have spirals in them, more like zig-zags.)
Or, you can try carbon composition instead of film resistors.

Jon

 

[George R. Gonzalez]

Thank you all for your suggestions-- I will try out most of them.

For the record, I was using 1/4 watt carbon comp resistors. I will try
again with 1/8 watters and see if that helps.
Better yet, I'll make up a resistor ringing test bench and try various
kinds. I can see how a spiral cut resistor could
be bad news at these frequencies!

For the output attenuator I had a 390 ohm resistor coming off the three
inverters, then a 51 ohm resistor to ground,
all with very short leads.

The bypass caps are wired directly on the IC, across pins 7 and 14 of
course.

The scope "probe" is a short piece of RG-58, next to no exposed wire, very
short shield soldered to pin 7.

Also could I reduce the resistor inductances by paralleling resistors? Or
does the increased pfds mostly negate the benefits?

 

[John Popelish]

Thank you all for your suggestions-- I will try out most of them. (snip)
For the output attenuator I had a 390 ohm resistor coming off the three
inverters, then a 51 ohm resistor to ground,
all with very short leads.

That should be 430 and 56 ohms, if you are shooting for a 50 ohm
output.

 

[John Larkin]

Thank you all for your suggestions-- I will try out most of them.

For the record, I was using 1/4 watt carbon comp resistors. I will try
again with 1/8 watters and see if that helps.
Better yet, I'll make up a resistor ringing test bench and try various
kinds. I can see how a spiral cut resistor could
be bad news at these frequencies!

For the output attenuator I had a 390 ohm resistor coming off the three
inverters, then a 51 ohm resistor to ground,
all with very short leads.

The bypass caps are wired directly on the IC, across pins 7 and 14 of
course.

The scope "probe" is a short piece of RG-58, next to no exposed wire, very
short shield soldered to pin 7.

Also could I reduce the resistor inductances by paralleling resistors? Or
does the increased pfds mostly negate the benefits?

If you're using carbon comps in the 1 ns timeframe, they're fine. Even
cheap carbon film resistors work OK at such "slow" speeds.

John

 

[Dana Raymond, a minor God]

Part numbers? I looked at the IC Master site and used google. Google gave a
lot of hits on 50 ohm ecl or lvds digital drivers. Try google: ""50 ohm" ecl
driver" and ""50 ohm" lvds driver". LVDS will give you lower bandwidth but
less exotic voltage swings than (P)ECL will. PECL operates easily into the
1-2 GHz range.

Hope this helps.
Dana.

 

[jmc]

Sooo I'd like to build a fast square wave generator to test out some
scopes.

I wire up a 74AC04, two inverters and a R/C are the oscillator, runs at
around 10MHz,
another inverter as a buffer, then three inverters in parallel as the
output stage.

Output goes to a 10-1 divider to match 50 ohms (and coincidentally to get
the voltage
down to the level the Tek smapling plug-in can handle)


All with very short leads, several bypass capacitors, ranging from 470pf to
20,000 pf.

It runs fine, putting out square waves with really fast rise and fall times,
faster than
my old Tek <400ps sampling plug-in can resolve. (75 ps plug-in on order).

PROBLEM IS, there's TONS of ringing on what should be the flat tops!

Adding bypasses helps SOME, but it mainly just changes the character of the
ringing,
doesnt reduce the amplitude by much.

Is this one of those counter-intuitive situations where "hard" bypassing
isnt what's needed,
I have to add a few ohms of series resistance on the bypasses? Weird
concept for sure.
How many ohms? Or should I just find some very small 50 ohm pots to put in
series with the various size bypasses and just tune everything for best
flatness? What a weird evironment.


Any hints appreciated,

George

Your layout is probably garbage. A 10mm loop has an inductance of about
10nH, which rings with a capacitance of 10nf (decoupling) at a frequency of
about 15MHz. Your best strategy is to use a ground plane or to at least have
the return trace running beside the source trace to minimize your
inductance.
jmc

 

[Chris Carlen]

Take a look at Fairchild and OnSemi single-gate parts. Many are
sub-ns. If you power an NC7NZ34 from +6 volts, and parallel all three
sections, it will put over 5 volts into 50 ohms in well under 1 ns,
and make a very pretty edge. And they're dirt cheap.


John

Usually I don't need to go so fast.

But the NC7NZ34 looks like it would make a good LED driver, too ;-)

Maybe I'll get some just to check them out.


--
_______________________________________________________________________
Christopher R. Carlen
Principal Laser/Optical Technologist
Sandia National Laboratories CA USA
[email protected] -- NOTE: Remove "BOGUS" from email address to reply.