Getting rid of the fuzzies

[Tim Williams]

So when you're working on an amp or waveform generator or whatever, what do
you do to clean up those residual cruddibits that tag along on your
precious waveform? Ringing, sloppy edges, the works. What's the trick,
ground plane, filtering, bypassing, pulse forming networks, ...!?

Right now I'm working on my vertical deflection amp project. I've soldered
down a pretty good circuit, with bandwidth out to 10MHz or so. (I had it
to 20MHz on the breadboard, where it was more stable. Yeah...) Having
discovered the CRT will deflect quite nicely with just 60V supply, I'm
having no problem with transistor selections.

Two problems remain. One is bandwidth: I had it out to 20MHz on the
breadboard, because it was stable with some emitter compensation on the
output stage (which is a cascode, 2N4401 into 2SC3597 running at 75mA). At
the moment, it isn't as stable (probably the lossy breadboard stray
capacitance, not to mention the ferrite beads I was using on things), so I
can't compensate the output stage any more without it turning into a
reasonable power oscillator at 30 or 100 or so MHz. So I need to look for
a solution to that. That will probably involve shoehorning in ferrite
beads.

The other concern, that I'm actually posting about: I get a straight
25-30ns rise time and, at low frequencies, perfectly flat tops. Woohoo.
But after that, there are about three small things wrong:
Right off the edge, there's some bouncing, to the tune of about 10%
(amplitude), at 20-30MHz or so, for about a microsecond. This appears to
be a complex mix of ringers. It looks weird on the (cheap?) CRT, almost to
the point of looping back on itself...
Over a somewhat longer scale, there's a slower bounce (about a MHz), of
suprisingly high Q: when tuning a sine wave through the range, the output
amplitude drops by about 10% over a narrow range of frequencies (Q ~
10-20?). After a step, it rings down in about 10us.

The easiest (so it would seem) to tackle should be the slower ring, which
seems like it would be a power supply thing. I tacked electrolytics onto
the high voltage supply (+40V) and that changed things a little, cutting it
in half. Now, I already have ceramic discs *everywhere* on this board, so
I'm not sure just how much more I can do in terms of bypassing. But I
can't think of a damn thing else that would give me this. And I've tried
poking ferrite rods around the circuit and nothing seems to change the
higher frequency bounce, either.

Tim

 

[BobW]

So when you're working on an amp or waveform generator or whatever, what
do
you do to clean up those residual cruddibits that tag along on your
precious waveform? Ringing, sloppy edges, the works. What's the trick,
ground plane, filtering, bypassing, pulse forming networks, ...!?

Right now I'm working on my vertical deflection amp project. I've
soldered
down a pretty good circuit, with bandwidth out to 10MHz or so. (I had it
to 20MHz on the breadboard, where it was more stable. Yeah...) Having
discovered the CRT will deflect quite nicely with just 60V supply, I'm
having no problem with transistor selections.

Two problems remain. One is bandwidth: I had it out to 20MHz on the
breadboard, because it was stable with some emitter compensation on the
output stage (which is a cascode, 2N4401 into 2SC3597 running at 75mA).
At
the moment, it isn't as stable (probably the lossy breadboard stray
capacitance, not to mention the ferrite beads I was using on things), so I
can't compensate the output stage any more without it turning into a
reasonable power oscillator at 30 or 100 or so MHz. So I need to look for
a solution to that. That will probably involve shoehorning in ferrite
beads.

The other concern, that I'm actually posting about: I get a straight
25-30ns rise time and, at low frequencies, perfectly flat tops. Woohoo.
But after that, there are about three small things wrong:
Right off the edge, there's some bouncing, to the tune of about 10%
(amplitude), at 20-30MHz or so, for about a microsecond. This appears to
be a complex mix of ringers. It looks weird on the (cheap?) CRT, almost
to
the point of looping back on itself...
Over a somewhat longer scale, there's a slower bounce (about a MHz), of
suprisingly high Q: when tuning a sine wave through the range, the output
amplitude drops by about 10% over a narrow range of frequencies (Q ~
10-20?). After a step, it rings down in about 10us.

The easiest (so it would seem) to tackle should be the slower ring, which
seems like it would be a power supply thing. I tacked electrolytics onto
the high voltage supply (+40V) and that changed things a little, cutting
it
in half. Now, I already have ceramic discs *everywhere* on this board, so
I'm not sure just how much more I can do in terms of bypassing. But I
can't think of a damn thing else that would give me this. And I've tried
poking ferrite rods around the circuit and nothing seems to change the
higher frequency bounce, either.

Tim

First you need to be sure that you're not seeing common-mode noise in the
scope. Connect the return side of the scope to where you would normally make
your measurement and then connect the probe tip to the return point, too.
Set the triggering appropriately, and If you still see the crap then it's
common-mode noise.

The most accurate way to do any measurement is with differential probes, but
not everyone has them and they have limited dc offset capability (although
you can externally cap couple to get rid of this limitation).

Bob

 

[Ken S. Tucker]

So when you're working on an amp or waveform generator or whatever, what do
you do to clean up those residual cruddibits that tag along on your
precious waveform? Ringing, sloppy edges, the works. What's the trick,
ground plane, filtering, bypassing, pulse forming networks, ...!?

Right now I'm working on my vertical deflection amp project. I've soldered
down a pretty good circuit, with bandwidth out to 10MHz or so. (I had it
to 20MHz on the breadboard, where it was more stable. Yeah...) Having
discovered the CRT will deflect quite nicely with just 60V supply, I'm
having no problem with transistor selections.

Two problems remain. One is bandwidth: I had it out to 20MHz on the
breadboard, because it was stable with some emitter compensation on the
output stage (which is a cascode, 2N4401 into 2SC3597 running at 75mA). At
the moment, it isn't as stable (probably the lossy breadboard stray
capacitance, not to mention the ferrite beads I was using on things), so I
can't compensate the output stage any more without it turning into a
reasonable power oscillator at 30 or 100 or so MHz. So I need to look for
a solution to that. That will probably involve shoehorning in ferrite
beads.

The other concern, that I'm actually posting about: I get a straight
25-30ns rise time and, at low frequencies, perfectly flat tops. Woohoo.
But after that, there are about three small things wrong:
Right off the edge, there's some bouncing, to the tune of about 10%
(amplitude), at 20-30MHz or so, for about a microsecond. This appears to
be a complex mix of ringers. It looks weird on the (cheap?) CRT, almost to
the point of looping back on itself...
Over a somewhat longer scale, there's a slower bounce (about a MHz), of
suprisingly high Q: when tuning a sine wave through the range, the output
amplitude drops by about 10% over a narrow range of frequencies (Q ~
10-20?). After a step, it rings down in about 10us.

The easiest (so it would seem) to tackle should be the slower ring, which
seems like it would be a power supply thing. I tacked electrolytics onto
the high voltage supply (+40V) and that changed things a little, cutting it
in half. Now, I already have ceramic discs *everywhere* on this board, so
I'm not sure just how much more I can do in terms of bypassing. But I
can't think of a damn thing else that would give me this. And I've tried
poking ferrite rods around the circuit and nothing seems to change the
higher frequency bounce, either.
Tim

As I understand it, the "ringing" is caused from circuit resonance,
specifically under-damped. It sounds like series resonance, so you
need some resistance in that series circuit to dampen it.
Of course you can tune out the dampening with a minimum
resistance, likely by that old standby, trial and error, or a pot
if your fancy.

Like hitting a bell, it rings, but if you grab it, it dies out
quickly.
Regards
Ken

 

[Ken S. Tucker]

First you need to be sure that you're not seeing common-mode noise in the
scope. Connect the return side of the scope to where you would normally make
your measurement and then connect the probe tip to the return point, too.
Set the triggering appropriately, and If you still see the crap then it's
common-mode noise.

The most accurate way to do any measurement is with differential probes, but
not everyone has them and they have limited dc offset capability (although
you can externally cap couple to get rid of this limitation).
Bob

I've had hassles going from protoboard to hard solder.
To avoid stray capacitance I use every other strip for
anything with high frequency, but also I've measured
contact resistance, which can bugger up (or improve)
a sensitive circuit.
I'm wild guessing, Tim's protoboard had enough wee
ohms to inhibit ringing.
Regards
Ken

 

[Tim Williams]

The most accurate way to do any measurement is with differential probes, but
not everyone has them and they have limited dc offset capability (although
you can externally cap couple to get rid of this limitation).

Ah, but I have a very good diff probe (well, as good as what's inside the
glass): a CRT! ;-)

Very good "DC offset capability", and AC CMRR, though focus and astig
change a bit in the process...

If you were referring to measurements like probing power supply points,
that's something entirely different. Hmm......but nah, the CRT isn't
sensitive enough to do that (10V/div). Hehe.

I know, I'll build another mostly-good deflection amp, this time
JFET-buffering both inputs to the diff amp stage, and probe around with
that!...

Hmm, come to think of it, a high bandwidth diff amp with wide CMRR (say,
30V) would be useful, and not too hard to build out of 2N3904's and
whatnot. Hmm, keeping dissipation down while keeping speed up would be a
problem.

Tim

 

[john jardine]

So when you're working on an amp or waveform generator or whatever, what do
you do to clean up those residual cruddibits that tag along on your
precious waveform? Ringing, sloppy edges, the works. What's the trick,
ground plane, filtering, bypassing, pulse forming networks, ...!?

Right now I'm working on my vertical deflection amp project. I've soldered
down a pretty good circuit, with bandwidth out to 10MHz or so. (I had it
to 20MHz on the breadboard, where it was more stable. Yeah...) Having
discovered the CRT will deflect quite nicely with just 60V supply, I'm
having no problem with transistor selections.

Two problems remain. One is bandwidth: I had it out to 20MHz on the
breadboard, because it was stable with some emitter compensation on the
output stage (which is a cascode, 2N4401 into 2SC3597 running at 75mA). At
the moment, it isn't as stable (probably the lossy breadboard stray
capacitance, not to mention the ferrite beads I was using on things), so I
can't compensate the output stage any more without it turning into a
reasonable power oscillator at 30 or 100 or so MHz. So I need to look for
a solution to that. That will probably involve shoehorning in ferrite
beads.

The other concern, that I'm actually posting about: I get a straight
25-30ns rise time and, at low frequencies, perfectly flat tops. Woohoo.
But after that, there are about three small things wrong:
Right off the edge, there's some bouncing, to the tune of about 10%
(amplitude), at 20-30MHz or so, for about a microsecond. This appears to
be a complex mix of ringers. It looks weird on the (cheap?) CRT, almost to
the point of looping back on itself...
Over a somewhat longer scale, there's a slower bounce (about a MHz), of
suprisingly high Q: when tuning a sine wave through the range, the output
amplitude drops by about 10% over a narrow range of frequencies (Q ~
10-20?). After a step, it rings down in about 10us.

The easiest (so it would seem) to tackle should be the slower ring, which
seems like it would be a power supply thing. I tacked electrolytics onto
the high voltage supply (+40V) and that changed things a little, cutting it
in half. Now, I already have ceramic discs *everywhere* on this board, so
I'm not sure just how much more I can do in terms of bypassing. But I
can't think of a damn thing else that would give me this. And I've tried
poking ferrite rods around the circuit and nothing seems to change the
higher frequency bounce, either.

Tim

The new layout is faster than the breadboard and fatally delayed higher
frequency components are again visible to the amp input. Fit the beads,
they really do help. Failing that, add shunt C somewhere to kill the HF
bandwidth. A ground plane will make the problem even worse.
Weirdly shaped ringing seems to turn up with mismatched transmission line.
Local LC ringing looks pretty. Tx line stuff looks to be composed of
parabolic curve segments.

 

[HapticZ]

copper plated grp board can add its own capacitance here and there, often
showing up as unwanted tuned ciruits where you have difficulty determining
where. even component leads can become tuned and generate harmonics

terminations at freq are critical to maintain stability and purity of
signals, as impedance matching is critical to retain maximum energy xfer
along the signal path, even if its as short as the distance fromi nput to
the output of an amp. consider the amp module itself as a 'line' with gain

proto/wire to grp is a challenge even for experienced designers since there
is so much change as you migrate

use standard via widths/spacings, plentiful ground planes, thru vias to
connect the grounds.

leave NOTHING to chance, or it will come back and bite you later like a
reflecting wave from the end of the line...