AoE x-Chapters, High-Speed op-amps section, DRAFT

There may be some generalizations that can be made, but there's a huge variation to be found from one op-amp to another.

On a sunny day (13 Jul 2019 14:02:59 -0700) it happened Winfield Hill wrote in :

I designed and build an audio mixer, used a very small inductor and cap at the inputs. ferrite bead might work too (have not tried that) against cellphones, there were none around back then. But we got audio from the electric tram radios that drove past the hall.

Filter any line level output too, to prevent any RF coming into the box. Metal box is a must.

That's exactly how it is. The noise attacks inside the chip at the first BJT pair so the closed loop gain no longer applies, it's at full open loop gain.

That is strange unless it had some sort of input protection that was really close to the Vf region of its diodes, such as back to back diodes between IN+ and IN=. Or if it was operated with an input very close to ground. However, AM stations are in the 1MHz range which is within the bandwidth of most opamps. Cell phones are >1GHz so the only way for that to get in is by rectification right at the beginning, not somewhere later.

The other thing to check for is whether the input stage was truly CMOS.

Yes, module level measures should always be first. It's just when they don't lead to success or seem too onerous to the client that I start changing components. Swapping in a CMOS opamp is usually an eye-popper to their engineers because it often completely eliminates the EMI. To my surprise none of the younger engineers ever knew this, or why. Often after explaining it to them they said "Why did our professors never talk about that?". That is a question I always had.

Rectified RF is no different from input DC offset. Both happen at the internal b-e junctions of the diff pair. But a unity-gain follower has Voffset at its output, not open-loop gain times Voffset.

So feedback magically doesn't apply? How exactly does that work? It works on every other source of offsets.

Cheers

Phil Hobbs

Well, a CFA with a FET front end would be much slower, other things being equal, because it would have only a few percent of the transconductance of a BJT.

Cheers

Phil Hobbs

The RF is added after the feedback divider. RF either comes in directly through a plastic enclosure or via power cables and the like and then re-radiated inside the box. The traces to IN+ and IN- form an unwanted dipole antenna, the RF couples onto those and then affects both BJTs in the first pair. It's like a comparator, the feedback loop is nearly powerless. While it does regulate out slower effects it cannot handle the sharp onset and drop-off of, for example, a GSM phone that seeks to establish communication with a cell tower. The pulses is what gets through and annoys, not DC.

It usually gets is after the feedback, see answer to Phil. The other problem is that the feedback is slow. DC and stuff is being regulated out but not the steep onset transients of a pulsating RF source. In an audio path it sonds lke "POCK-POCK-POCK".

The fix is easy but the downside is that CMOS opamps have much worse input voltage noise and also often cost more.

So is DC offset.

RF either comes in directly

If the forward gain of the opamp is slow, it's slow for both the RF-induced offset and for the feedback.

We could Spice that, but we'd have to decide where inside the loop to inject the RF offset.

Yes, but it does not pulsate. The transitions are the problem. If it were a constant RF carrier there would be not a problem other than maybe an elevated noise level.

Immediately at (inside) the BE junctions of the first pair. It's hard to do because RF also attacks other BE junctions in there.

The problem is that such signals develop inside the opamp and the feedback loop only notices it after the fact. "Oh s..t! Where did that come from? Let's fight it!".

Okay, so you're talking about rectified components near or beyond the loop bandwidth. I agree with that--there's not so much difference between A_VOL and A_VCL up there. But that's because A_VOL isn't 100 dB up there either.

Down at the 220 Hz GSM data rate, your average op amp has plenty of speed to respond to rectified RF in the approved closed-loop manner.

Cheers

Phil Hobbs

When I used to work on Judah Street, I drove to work over a road on Twin Peaks, just under the Sutro tower. My old ratty Fiesta had a cheap add-on radio. I could hear buzzing from the speakers with the radio turned off.

People live up there.

I was first "hit" almost literally by GSM RFI in ~1994. Someone had a newfangled digital phone, and the ba-da-da ba-da-da ba-da-da bzzzz was *very* loud in my deafaid.

I almost dropped my coffee cup, and if I had been holding an infant, I could have dropped that :(

I did a little research and measurement, and found that DCS1800 was much worse than GSM900, and that the audio power dropped off as a function of distance somewhere between r^4 and r^8, and was proportional to f^2(?) up to at least ~3GHz.

That put me in an awkward position at work; I was lucky that the company was a responsible company in many ways.

My experience with opamps rectifying RF on PC boards is that there are usually several narrow resonances where it's very sensitive, in the low 100s of MHz.

One of my (ex) competitors NMR temperature controllers, with a thermocouple sensor, could be hard shut down with a GR signal generator, from clear across the room.

An RF design friend of mine has had issues with parallel capacitors (like 0.1uF/100pF) adjacent on a supply line. He's had quite high-Q resonance between the 100pF and the parasitic inductance at between 400 and 900MHz. Traps, and he's not a "young player".

In a previous job, we had a lot of interference from a 2.4GHz ISM band frequency hopping transmitter with a 5ms TDMA slot. We were shipping for about a year without any problems, then it hit the fan. Turns out that the opamp vendor changed the opamp process and increased the transistor Ft. We spent the better part of a year trying to get rid of the 200Hz buzz reliably.

On a sunny day (Mon, 15 Jul 2019 16:41:44 -0700) it happened John Larkin wrote in :

It is common, my Sony alarm clock FM radio also makes noise when my cellphone interacts with the tower, even with that radio off, so do my PC speakers.

I didn't spot any and doubt there were any, but I wasn't looking for them. Other effects were of more interest.

With deafaids - there are minimal components so fewer opportunities for resonance and - the circuits are much smaller than typical PCBs, so antennas are "short" and not operating efficiently near a resonance.

People who have done EMI/EMC studies are confused that the coupled RF power is theoretically proportional to f^2. Their assumption is that the power falls with increasing frequency. The difference is that they are concerned with worst-case susceptibility with the most efficient possible RF coupling. That isn't relevant in this case.

Looking back at my notes...

Even in an anechoic chamber physical movements caused 5dB changes in coupled RF power (corresponds to 10dB audio)

The coupled RF power at 3GHz was 7dB more than at 1GHz, corresponding to a f^1.2 dependency, not the theoretical f^2 dependency. That was supported by other ETSI studies.

The audio power is proportional to the square of the RF power.

Overall the audio power was proportional to - tx power ^2 - f ^2.4 (ish) - 1/distance^4 to 1/distance^8

I've just noticed I have another easily accessible result...

For one type of deafaid, the audio power fell off rapidly above 2.2GHz, for others it was above 3.5GHz and not so rapid.

The cause was not investigated, but obvious guesses could be made.