Super TGSL With H Bridge TX Diff RX Quad Sampling

[Davor]

It is odd. Is the waveform the same as before?

[Silver Dollar]

Yes the same. It seems like there is no hysteresis built in. It oscillates at 4 mhz...

[Davor]

Can you see what is going on at the inputs of the comparator? Maybe a little hysteresis will fix it without introducing any other artefacts. I did not anticipate any instability there because of relatively large input signal and relatively low input impedance.

[Gwil]

Oscillation is sometimes due to insufficient decoupling of the power supply rails. I used to work on aircraft systems in which every op amp package had its own 0.1uF ceramic capacitors placed as close as possible to it.

[Davor]

That's correct, however, comparators in quad packs often show glitches as a consequence of mutual coupling of some kind. These are not problematic if every I/O is on low to moderate impedances.
I believe Silver Dollar decoupled these, but I inquired about what's going on at the negative rail, as that one is a bit more critical. One of these days I'll revise the circuit to include a bit of hysteresis. Most probably it will fix it for good.

[Silver Dollar]

I'm still trying to get this going. As it stands there is only the oscillator and a LM393 in the circuit.
I have decoupled the supply's and even used batteries with no change. A funny thing, the signal
gets noisier the further along the signal path it goes and there is no noise source around. Granted
we are attenuating it a bit. I have tried using fixed resistors instead of the pots and looking at it
today (after adding a bit more decoupling) it sort of works. At one end of the pot I get a square
wave but it becomes an envelope with 1.5 MHZ oscillation it it at the other end of the pot.

I made a drawing of what signals I see. It's very small and noisy on the comparitor side of the
68K's;

[Davor]

In Bosnia there is a saying: hrkljuš (see http://www.youtube.com/watch?v=u0UqLsCCIv8 )
When deciding values I overdid it in a direction that resulted in somewhat small-ish resulting comparator input signal. I have no idea where the input branch picks that oscillation, but if given a bit more juice it will surely be less of a problem.
I'll correct the values and post the changed schematic here. There will be very few changes, don't worry about it. The linearity of a transfer function will not suffer significantly (which is the fact I had to realise sooner). The whole idea was to supply a clever arrangement of potentiometers with orthogonal signals at proper amplitude ratios, so that resulting phase transfer function goes linearly with potentiometer position. A wincy bit less orthogonal, but at a significantly larger amplitude should do the trick.

[Silver Dollar]

That's what I was thinking when I saw the 50mv. I just wasn't sure how to tweak it without messing up the outputs.

I had a chat with an engineer at my new job today and he suggested some filtering on the feeds to keep the comparator
happy....

[Davor]

Well, it must be well decoupled. Trouble is that comparators within a chip tend to interfere one another if each chip is not properly decoupled at shortest possible proximity. It happens.

I tweaked the values and concluded that there is a little difference phase wise, but amplitudes are some 4 times higher. The phasing network is corrected at R11 and R12, C22 and C23. C10 and C19 can be reduced, but also may remain as they are. My mistake was in a fact that the signal from the phasing network never drops below some minimum which is close to a maximum, and I simply forgot to check the minimum value. (hrkljuš)
For extra stability it is possible to add Rhyst and Chyst at 5pF and 1Mohm. I'll try these in my IGSL and see if these may improve stability. It makes sense to use hysteresis to remove uncertainty, as there is a bias at the switchers that turns every uncertainty into noise.

[Silver Dollar]

Still some trouble with oscillations. I changes the 220 ohms to 1K and took some pics;






Not too bad but a little messy. So I added the 5pf and 1M's and it got worst;




I have a .1 and .01 right on the power pins of the chip and 20uf and 1uf and .1's on the power supply.
The chip is in a socket. It's a machined pin gold plated socket but maybe it's corodded inside the holes?

[Davor]

I see the comparator output is about 10V. What kind of stabilisation is there? The comparators are supposed to be supplied from 8V stabilised and a signal ground - not minus 5 volts. Negative supply is not required for them as switchers are 4053 that have Vss and Vee pins and support a shifted level switching. I have no idea what would happen if the comparator supplies 4053 with levels below 0. Only Vee of 4053 is supposed to go to minus 5V.

This oscillation business makes very little sense, but heck, it is going somewhere. Did you change 100n to 22n as well? Both 220ohm and 100n must be replaced to keep the same phase constellation.

A positive feedback was supposed to wipe oscillations, and not promote them, so please check if maybe the feedback was connected to the non-inverting input by mistake. Such connection would surely make it oscillate.

In case there is some external pickup of these oscillations, it is OK to place a 47pF capacitor between the inverting and a non-inverting input of a comparator. It would shift phase for additional 3° or so - no big deal.

I guess there is a problem with lead lengths in the phasing network. It is spring and it means I may reach my lab hut and try something. If you observe the top of the comparator output (logical 1), you'll notice a small "hairy" patch in a middle which is a consequence of the other channel going through transition(s) and that happens regardless of decoupling and becomes very bad if there is no decoupling. You can move the hairy patch left and right by shifting the potentiometer of the other channel. There is no way of making it disappear completely regardless of decoupling, but it is not a problem.

[Silver Dollar]

Well I'm using +-9V temporarily. The comparitor is wired to +9 and gnd. I did notice the inputs
to the comparitor were centered on ground (IE going above and below ground). I thought they
were at +4.5V before - I'll have to double check that.

I'm still using a LM393 and attached the 5pf and 1Meg to pins 1 and 2 and to pins 6 and 7.

I didn't change the 100n's yet, I'll try that tonight...

I didn't attempt to make the leads short but they are not super long either. Yes it's a strange
problem.

[Davor]

OK, one problem is solved then. Feedback must go between 1 and 3 and also 5 and 7. You are after a positive feedback which in this case turns into hysteresis and makes oscillations go away.
If there will be any wander (jitter) in comparators output left after this feedback thing is applied, I'll see to increase amplitude even further, but I doubt it will be necessary.

[Davor]

One more thing ... the comparators inputs are seeing impedances in a range of ~20kohm and input signal is not too high. In case there is any conductivity in between the pins at all, say in vicinity of 1Mohm, and because the inverting inputs are next to the outputs, you'd have a negative feedback path and oscillations as a consequence.
Gary (Chemelec) often warns about this conduction problem, and his solution is using a water soluble flux cored solder (!) and he washes his boards in water. I tend to prefer resin flux (smells better too ) and rub the critical parts with alcohol. It is very bad to use a water soluble flux and leave it there because it is conductive.
Either way, I believe the hysteresis will fix this.

[Silver Dollar]

Well some success today. The bare boards I use come coated in flux to solder easily.
I tried to remove the flux with alcohol and scraping with an exacto but wasn't sure how
clean it was so I brought it to work where we have flux-off and some ultrasonic cleaners.
I cleaned the board and retested;



Maybe a little better? Then I added the hysterisis R+C;



One looks good the other is ok in the middle of the pot but
oscillates at each end. I haven't changed the 100n caps yet
I'll try that next...