Is anybody using NE521 comparators?

How do you think we can know without you showing us exactly what you did?

Jeroen Belleman

Are you looking at output with a scope- could be oscillating?

Getting it to oscillate faster than your oscilloscope can follow?

The NE521 is quite fast - it can get up to 55MHz - and it's not difficult to make it self-oscillate if the voltage difference between the input isn't high enough to swamp capacitative feed-through or ground loops.

I've had to crawl through circuits which were in production when a batch of faster comparators - not NE521 parts - showed up a marginal layout. And tightening up the layout didn't always keep on working.

If the strobe pins are high, the output will be undefined.

That's not so. The NE521 has no latch of flipflop. What they call strobes are actually inputs to NAND gates. If any of the strobes are low, the associated output(s) will be forced high.

The output is a traditional Schottky TTL gate. If the OP sees something in between a definite high or low, it's probably oscillating. That could be because one or both of the comparator inputs are driven from a too-high impedance, or it could be feedback due to poor layout, or it could be poor or missing power supply decoupling, or something else yet.

In summary, let's wait until he shows us what he did.

Jeroen Belleman

P.S. Half of the equivalent schematic in the datasheet is nonsense.

Yes this was an oscillation. A bit fast for my scope but visible.

I worked around with a 10 mV hysteresis but it is unpleasant because I want to detect when an about 20 mV high signal goes to zero. I oscillates with a smaller hysteresis.

If you've got control over the circuit layout, you can look for the likely feedback paths and try to reduce them - if the printed circuit layout was sub-contracted, this can sometimes be quite easy.

I remember an occasion from early in my career - at EMI Central Research

- where we got given a layout by a subsidiary company and could see - by inspection - that a particular amplifier chip was going to oscillate.

We could rotate the chip by 180 degrees and separate the input track from the output track, which made the power supply tracking a bit messier, but turned out to solve the problem.

The subsidiary had already made and loaded 210 of the boards, and refused to junk them, choosing instead to take out the chip (which did indeed oscillate) and bodge in a two transistor amplifier, which must have cost a fortune.

Crankier problems can call for extra ground track between sensitive tracks to provide a measure of capacitative shielding.

On one occasion I ended up soldering a length of heavy copper wire onto a ground trace, which lowered it resistance enough to kill off a nasty ground loop.

The production modification was less messy - we split the track so that the offending high current went down one track while the other track served as OV voltage reference - "star earthing".

Perhaps I should have prefaced my reply: "The data sheet says" if the strobe pins are high...

"offset" has a particular meaning for its designers. Tha datasheet says: output undefined if abs(Va_Vb) < offset. Forget what you know about the offset of an opamp! One could expect better from a bipolar differential pair :-)

Well, you could improve the layout, or add an amplifier in front. How fast do you need it to be?

Cheers

Phil Hobbs

What's wrong with that?

The offset for any particular pair of nominally identical transistors can be more tightly defined than that, though it can be expected to drift with temperature, but this is a data sheet talking about all the possible parts that the manufacturer might sell under that part number.

If the two inputs are closer in voltage than the data sheet value for the maximum offset, they can't say anything reliable about the state of the output.

It can be seen as a cop-out, but it's hard to see how they could be more specific.

It's Va-Vb, the difference. Not Va_Vb.

They mean that they can't tell if the output will be either high or low; Not that it is hovering somewhere in between. This is pretty much what one should expect from a comparator offset. Of course, for *really* small inputs, the output will start wiggling randomly. You're not supposed to go there.

I agree Vos=6mV is pretty poor for a bipolar differential comparator, but hey, this is an old design. It can easily compare voltages less than the offset, but you'd have to get serious about correcting for it. It gets slower, too.

Jeroen Belleman