Whining noise from circuit board

[richard.cs]

Ceramic capacitors with a deliberately added series resistor are a perfectly legitimate alternative to electrolytics for regulators that require them. If you have board space it's often a good idea to put a 0R in series with the output capacitor anyway and then the footprint is there if you have problems. Bear in mind that a controlled ESR output capacitor might not be enough if there are ceramic decoupling capacitors sprinkled around elsewhere on that net - the regulator will still see their low ESR unless they are at the ends of long inductive traces (which can cause other problems).

[Asmyldof]

I feel like chiming in, because the cause has been pointed out, but I think maybe an explanation might help future design thoughts.

What happens with a linear regulator of a certain feedback design and a "too perfect" capacitor is that the regulator has an internal structure like an op-amp, with some inherent hysteresis and offset in its measurement inputs (because no op-amp you can buy at $1, let alone one with a power stage, is designed for near-perfect behaviour).

So, let's say the output starts at 0V, the output stage of the regulator starts pumping out power to fill the cap. Because of the offsets and some "significant" delay/phase-shift it needs some time to "get used to" a new voltage measured over the cap.

A cap with very low ESR on an output stage that will happily supply 100's of mAmps shoots up at great speed. So the regulator then 'sees' 5V a bit after it achieved this, with the great response of the low ESR cap, this can mean 5.5V is actually reached when it starts pinching off. Now, the cap needs to drain to 4.99V before it starts opening up again.
(voltages assume the regulator to be exactly 5.00V - to the concept the variations don't matter)

This creates the oscillation.

What's worse than there being a sound, is that this means that the internal power transistor is continuously switched on and then pinched off, switched on and pinched off. This transistor is designed for use that's constantly in its linear operation, as a regulator expects to always source a little bit of current at least. That's why some regulators even specify a minimum of 5mA drain or something like that.
Now using this transistor as a sort of switching type may cause it to become quite a bit hotter than you intend, since its doing stuff it's not designed for.

Some linear regulators have an internal structure that includes a cascode power stage as a complementary buffer resistance, or include a special feed-forward structure, that allow you to go as low as single mOhms ESR, but these can primarily be identified by their >$5 price tag when bought in volume.
And of course most switching bucks prefer low ESR as they actually work with a current dump-scheme and switching transistors.

[hpmaxim]

Ceramic capacitors with a deliberately added series resistor are a perfectly legitimate alternative to electrolytics for regulators that require them. If you have board space it's often a good idea to put a 0R in series with the output capacitor anyway and then the footprint is there if you have problems. Bear in mind that a controlled ESR output capacitor might not be enough if there are ceramic decoupling capacitors sprinkled around elsewhere on that net - the regulator will still see their low ESR unless they are at the ends of long inductive traces (which can cause other problems).

There are several .1uF caps spread around as decoupling caps, which was why I had said no output cap was needed in response to Wraper, I could simply rely on those.  My general assumption is that ceramic capacitors will last longer than electrolytics, particularly in harsh environments which is why I went with them.  However, not needing the 10uF caps, it looks like I can get a better lower capacitance substitute which will also be significantly cheaper.

[wraper]

There are several .1uF caps spread around as decoupling caps, which was why I had said no output cap was needed in response to Wraper, I could simply rely on those.

You can rely on those only if they are located very close to the voltage regulator IC. Decoupling capacitors across the board are not a substitute for it.

[hpmaxim]

I feel like chiming in, because the cause has been pointed out, but I think maybe an explanation might help future design thoughts.

What happens with a linear regulator of a certain feedback design and a "too perfect" capacitor is that the regulator has an internal structure like an op-amp, with some inherent hysteresis and offset in its measurement inputs (because no op-amp you can buy at $1, let alone one with a power stage, is designed for near-perfect behaviour).

So, let's say the output starts at 0V, the output stage of the regulator starts pumping out power to fill the cap. Because of the offsets and some "significant" delay/phase-shift it needs some time to "get used to" a new voltage measured over the cap.

A cap with very low ESR on an output stage that will happily supply 100's of mAmps shoots up at great speed. So the regulator then 'sees' 5V a bit after it achieved this, with the great response of the low ESR cap, this can mean 5.5V is actually reached when it starts pinching off. Now, the cap needs to drain to 4.99V before it starts opening up again.
(voltages assume the regulator to be exactly 5.00V - to the concept the variations don't matter)

This creates the oscillation.

What's worse than there being a sound, is that this means that the internal power transistor is continuously switched on and then pinched off, switched on and pinched off. This transistor is designed for use that's constantly in its linear operation, as a regulator expects to always source a little bit of current at least. That's why some regulators even specify a minimum of 5mA drain or something like that.
Now using this transistor as a sort of switching type may cause it to become quite a bit hotter than you intend, since its doing stuff it's not designed for.

Some linear regulators have an internal structure that includes a cascode power stage as a complementary buffer resistance, or include a special feed-forward structure, that allow you to go as low as single mOhms ESR, but these can primarily be identified by their >$5 price tag when bought in volume.
And of course most switching bucks prefer low ESR as they actually work with a current dump-scheme and switching transistors.

Again, I had my blinders on because I thought it was just a linear regulator, with a similar architecture to the 7805.  However, after looking at the block diagram, I noticed they are using a PNP for their output power transistor.  This is almost a dead giveaway that they can't be operating in emitter follower/linear.   I'm a little surprised the part is designed like this.  They are essentially relying on the load being well matched to the speed of the regulator -- although I guess that's the price you pay to get close to the rail.  I actually designed a current current sink/source IC that had to work close to the rail and ran into the same sort of issue.  However, I vociferously warned the customer of all the potential issues with oscillation because they were unable to guarantee the load (they were making a box which was being sold to their customer who in turn could attach whatever they wanted to the outputs).  The customer accepted this because they knew it was the only architecture that could meet all their specifications.  But it would be nice if they gave you a bit more warning in the datasheet.

I don't buy the argument that would cause it to heat up excessively.  Oscillation at the base could cause you to burn through energy to charge capacitance, but I can't imagine it'd be more than 1mW (I'm figuring 1/2 * 1nF * Vin^2), and considering that they are speccing 6 mW quiescent when nothing is happening, going up to 144mW under load, and you're inherently losing (Vout-Vin) * current anyway, I doubt the extra 1mW is going to do much. 

In all honesty, I suspect this entire situation is tolerable, unless the physical oscillations would eventually cause the solder joints on the cap to crack.  I also worry that given the frequency and amplitude this could adversely affect the operation of the analog ICs, although I suspect its slow enough it wouldn't affect the digital.  But given that I can fix it by selecting different parts which are actually cheaper and better able to handle high voltage, that's the obvious solution.

BTW, as I recall from the design I had worked on.  18V is the maximum voltage you ever expected to see in a functioning or semi-functioning car, although you needed tolerance to about 36V because of abnormal events (someone jump starting the car by somehow wiring the batteries in series, and then getting a load-dump (the alternator becoming disconnected from the battery or something goofy like that.

[paulie]

18V is the maximum voltage you ever expected to see in a functioning or semi-functioning car,

I have personally measured 160v in a 1998 Dodge Caravan and somewhat more in a 1984 Fiero under normal operating conditions. I suspect significantly higher spikes exist of shorter duration.

[hpmaxim]

18V is the maximum voltage you ever expected to see in a functioning or semi-functioning car,

I have personally measured 160v in a 1998 Dodge Caravan and somewhat more in a 1984 Fiero under normal operating conditions. I suspect significantly higher spikes exist of shorter duration.

Yeah, I don't believe it.  I guarantee you that most electronics in a car would be destroyed by that, and in a very short period of time.  I know this because we were destroying parts during ESD testing with 200V (machine model, not  HBM).  And the customers (two of the largest automotive electronics manufacturers in the world) had no concerns about this, and neither did their customer (one of the top 5 automakers in the world).  And any prolonged exposure above about 44V would destroy it (prolonged being a second or less) -- and they understood this as we blew up a lot of parts while doing testing.

I strongly suspect something was wrong with your measurement setup.  The automotive environment is very dirty, a bad ground connection could result in picking up major league noise that isn't really there.  Now, if you mean something like a inductive pickup, its possible you could get large voltage off that, especially if its disconnected from the interface circuitry -- however the interface circuitry almost certainly has either some shunt resistor which would reduce the loaded voltage, or some sort of series resistor and a shunt cap to reduce peak voltages at the interface IC for high frequency (and hence high amplitude) input signals. 

[Asmyldof]

I don't buy the argument that would cause it to heat up excessively.  Oscillation at the base could cause you to burn through energy to charge capacitance, but I can't imagine it'd be more than 1mW (I'm figuring 1/2 * 1nF * Vin^2), and considering that they are speccing 6 mW quiescent when nothing is happening, going up to 144mW under load, and you're inherently losing (Vout-Vin) * current anyway, I doubt the extra 1mW is going to do much. 

In all honesty, I suspect this entire situation is tolerable, unless the physical oscillations would eventually cause the solder joints on the cap to crack.  I also worry that given the frequency and amplitude this could adversely affect the operation of the analog ICs, although I suspect its slow enough it wouldn't affect the digital.  But given that I can fix it by selecting different parts which are actually cheaper and better able to handle high voltage, that's the obvious solution.

In addition to Wraper's graphical proposition to you, which is entirely a second very valid point, you seem to be misunderstanding what I'm explaining.
A transistor made for continuous conduction is , 9 out of 10 not good at switching and vice versa, no matter its type or polarity. They are optimized for a task. It's like taking your 1.6liter station wagon to a drag race and then using your Ferrari to do a monthly shopping run, it's just not the best idea.

A transistor designed for linear operation is going to detest switching and waste a lot of energy doing so, even though half that energy might have otherwise gone into the cap making the oscillation worse.
Also, not just one transistor that wasn't meant to is oscillating between off and on, the entire chip is!

Add that to the large AC currents that could be leaking as Wraper rightly points out, and, pfffooah as a locally famous Ozzie might say.

So, I very much think this is not a manageable situation, even if the heat stays in limits and the sound can be masked or hidden, you are introducing operation to several parts that they were absolutely never designed for and certainly never tested in. To go back to the story of your Ferrari, even if it works while you try it, at some point something's going to give under the weight and packing conditions. (Or if you think of this as a station wagon, it'll probably cost you your clutch, transmission, head gasket or all three in the long run).

[wraper]

I deleted the post Asmyldof says about after a minute since posting (before he made his post) as it wasn't 100% accurate. Just don't wonder what he says about.

[paulie]

Yeah, I don't believe it.  I guarantee you that most electronics in a car would be destroyed by that, and in a very short period of time.

Obviously not if designed with proper protection by a competent engineer.

You are free to continue with head in the sand. Sometimes what you don't know CAN hurt you. Other times survival may continue due to dumb luck. The fact you mention "a second" tells me you have no idea what I am referring to. Of course you are entitled to continue with your belief system. It's just as amusing to see those who live in blissful ignorance as it is to see the ones who take ridiculously excessive measures. That automotive protection megathread (largest ever in this forum), with both extremes, was truly a comic opera.

BTW with soldering iron in hand it would take me about 10 minutes to isolate the cause of a noisy component. Much less time than spent on this thread anyway. If you pull the legs off an ant he is at some point unable to continue crawling.

[dom0]

18V is the maximum voltage you ever expected to see in a functioning or semi-functioning car,

I have personally measured 160v in a 1998 Dodge Caravan and somewhat more in a 1984 Fiero under normal operating conditions. I suspect significantly higher spikes exist of shorter duration.

Yeah, I don't believe it.  I guarantee you that most electronics in a car would be destroyed by that, and in a very short period of time.  I know this because we were destroying parts during ESD testing with 200V (machine model, not  HBM).  And the customers (two of the largest automotive electronics manufacturers in the world) had no concerns about this, and neither did their customer (one of the top 5 automakers in the world).  And any prolonged exposure above about 44V would destroy it (prolonged being a second or less) -- and they understood this as we blew up a lot of parts while doing testing.

There's some standards for testing automotive stuff around, I think common criteria are something like "must survive <1 ms pulse with five (seven?) times the normal voltage" and "must survive incorrect polarity".

/edit: http://www.dse-faq.elektronik-kompendium.de/dse-faq.htm#F.23

     Test Level I/II/III/IV according to ISO 7637-1:
    Pulse 1: -25/50/75/100V, rise time 1us, width 2ms, source resistance 10
    Ohm, repetition rate 0,5 bis 5s, test duration 5000 Impulse.
     Pulse 2: + 25/50/75 / 100V, otherwise as pulse 1
     Pulses 3a and 3b: pulse packets of pulses -25 / 50/100 / 150V (3a) or
     + 25/50/75 / 100V (3b), with rise time of 5 ns, duration 0,1us, internal resistance
     50 Ohm, repetition 100us, packet duration 10ms, 90ms interval between packets,
     1 hour test period.
     Pulse 4: Sagging of the voltage by -4 / 5/6 / 7V about 0.1 seconds and -2.5 / 6V to
     20 seconds.
     Pulse 5: 40-400ms long pulse of 26.5 / 46.5 / 66.5 / 86.5V with 0.1-10ms
     Rise time, internal resistance 0.5-4Ohm
     JumpStart / starting aid: several minutes 24V (or more precisely 28.8V) for 60sec
     Charging: 17V for 60min

[paulie]

Sometimes what you don't know CAN hurt you.

Anybody get the pun? You know... "CAN" bus... automotive... LOL.... never mind...

[hpmaxim]

Yeah, I don't believe it.  I guarantee you that most electronics in a car would be destroyed by that, and in a very short period of time.

Obviously not if designed with proper protection by a competent engineer.

You are free to continue with head in the sand. Sometimes what you don't know CAN hurt you. Other times survival may continue due to dumb luck. The fact you mention "a second" tells me you have no idea what I am referring to. Of course you are entitled to continue with your belief system. It's just as amusing to see those who live in blissful ignorance as it is to see the ones who take ridiculously excessive measures. That automotive protection megathread (largest ever in this forum), with both extremes, was truly a comic opera.

BTW with soldering iron in hand it would take me about 10 minutes to isolate the cause of a noisy component. Much less time than spent on this thread anyway. If you pull the legs off an ant he is at some point unable to continue crawling.

First off, I've already acknowledged that I selected a suboptimal regulator without realizing it was a totally different architecture, and intend to replace it on the next board that I build up.  I'm not convinced the ringing is likely to cause long term damage, but it is mildly annoying, and it could affect the analog ICs, and there are better/cheaper alternatives. 

Second, I have no interest in getting into a pissing match about this.  The "competent engineer" comment is really out of line.  I do not believe that most automotive electronics are designed to survive more than 45V, certainly not statically.   GM, Ford, Toyota, VW etc, all have specifications which have been agreed upon as to what the electronics are designed to survive, and you are expected to meet that and not more.  I was designing an IC that went into a module, its quite possible that there was additional circuit protection in the module to protect the IC against spikes (we were not privy to the customer's requirements from the automaker, only the derived requirements they gave us).   We interpreted the 40V specification to be both a worst case static and transient case, however it was probably only intended as a transient case.   It's conceivable to me that there were TVS diodes in the module which would protect the IC from transients above 40V and I may be wrong about that, but I have a lot of trouble believing they have transients going up to 160V, because if there is TVS protection, the modules should unless isolated by a resistor or something be protecting each other at least to some degree (admittedly, for very very fast transients, probably not as the inductive impedance might create substantial isolation).  When you mentioned the 160V there was no mention of how long the spike was versus a static voltage level, but you implied that for "shorter durations" you believed there were larger spikes.

Anyway, I do appreciate the feedback, it appears the problem has been determined, the solution has been determined and I learned stuff in the process (hopefully others did as well).

[paulie]

A few minutes reviewing your posts don't see any hint the problem has actually been determined. Maybe you forgot to mention the details of your find.

Anyway it's good to hear about the possibility of addition protection in the module your circuit is used in. All this talk of hundred volt spikes is not as scary as might seem first glance. For personal automotive projects nothing more than a single resistor on inputs has served me fine for decades. Both in the vehicles mentioned above and also some 2 wheelers. Feels strange to be on the other side of these "pissing contests" for a change.

BTW you seem somewhat overly sensitive based on replies to other contributors on the first page too. The "competent engineer" remark was not directed to anyone in particular.

[hpmaxim]

A few minutes reviewing your posts don't see any hint the problem has actually been determined. Maybe you forgot to mention the details of your find.

Anyway it's good to hear about the possibility of addition protection in the module your circuit is used in. All this talk of hundred volt spikes is not as scary as might seem first glance. For personal automotive projects nothing more than a single resistor on inputs has served me fine for decades. Both in the vehicles mentioned above and also some 2 wheelers. Feels strange to be on the other side of these "pissing contests" for a change.

BTW you seem somewhat overly sensitive based on replies to other contributors on the first page too. The "competent engineer" remark was not directed to anyone in particular.

Well, I determined that the regulator I was using did not have an emitter follower output (making it more susceptible to output oscillation) and was clearly oscillating at 14-15kHz which was probably the frequency I was hearing.  The solution would be either to add series resistance to the ceramic capacitor or (the preferred solution is to) change to an emitter follower based regulator such as a 7805 which would be unconditionally stable.

I think you may be confusing two different things...  The module I was talking about was a module that I designed an integrated circuit for at work.  Again, the specifications that I was given were extremely tough to meet in an IC, but they were almost certainly derived specifications, i.e. their customer had given them specifications for the module and based on PCB circuitry they made a a decision as to what IC could be exposed to, which would be less than the module's exposure.

After looking at the schematic of the PCB I'm designing (for fun, not for work), I'm thinking my protection is probably inadequate.  The inputs are properly protected, but the outputs are directly connected to automotive-rated IC.  That's probably not adequate, throwing in 100 ohm series resistors.  My power supply is connected to the LF50ABDT through a rectifier diode.  Now, I'm thinking the diode should be replaced by a 10 ohm resistor with a shunting TVS.

[Asmyldof]

I deleted the post Asmyldof says about after a minute since posting (before he made his post) as it wasn't 100% accurate. Just don't wonder what he says about.

It wasn't 100% accurate, but still a valid point that an AC wave on a cap will cost you some that's not measurable past the cap.

but I have a lot of trouble believing they have transients going up to 160V, because if there is TVS protection, the modules should unless isolated by a resistor or something be protecting each other at least to some degree (admittedly, for very very fast transients, probably not as the inductive impedance might create substantial isolation).

I have no trouble at all believing in a car with even 100's of tested devices the main lines still contain spikes past the 100V point.
I also have no trouble believing a small spike like those can destabilise the input stages of a voltage regulator specced to an absolute maximum of anything below 60V (and even with 60V+ ... protection)

The module test is very nicely quoted by dom0 and as you can see the tests do go up to 150V.

Although the total energy is very low and a simple R-C filter (low R, high C) can usually flatten them off to below 50V, while still allowing sufficient primary current to flow, if you worry about inductances and after-dumps a few more components could be added, such as a TVS and diodes. It will not pass a professional test for production quality modules, but it'll do the job well enough for a one-off.

Also don't be fooled by personal measurements, especially when taken with a digital scope or some similar such. A 160V 0.1us spike can very well not show up at all on your display, or show as a simple 30-ish spike of 0.05 ~ 0.5us. The sampling has its limits and even on analogue scopes the trace of the spike can be so thin that at first glance you miss it entirely. I have spent enough hours in nearly completely dark rooms twisting focus and brightness knobs just to be sure in similar tests and measurements.

As for single stage emitter-follower outputs never oscillating. Yes they sometimes do. It's slightly harder to get them to, but the op-amp stage of a linear regulator will still introduce phase shifting which leads to rush-in-based overcharge and pinch off followed by depletion, turn-on, over-charge, pinch-off, depletion, turn-on, etc. If the C is dampened by a resistance (internal or external) these peaks get smothered to a degree that'll stay compatible with the phase-shift/delay in the error amplification and the oscillation will reduce to (near) nil.
The key in non-oscillation with low ESR capacitances is in the main control circuitry's cleverness with regards to that. (compensation caps on extra pins, feed-forward stabilisations, whatyamaevers)
Sometimes this cleverness is a side product of another intended quality, sometimes it is fully intentional.

With 7805 types, most of them are of rugged enough design (case of luck or research, I do not know, but it's one reason they have been popular for ages) to be very hard to get to oscillate in any significant way. But careful examination of the specific device's (manufacturer dependent) datasheet is never misplaced with regards to ESR requirements.

[hpmaxim]

I might as well ask about the protection circuitry that is employed, since its clear I don't really have experience designing automotive circuits:

I'm protecting the 7805 with two 47 ohm series resistors (.75W each), 25V TVS shunt, 1uF 50V shunt cap.
I'm protecting digital inputs with a 1k  series resistor (.4W), 5V TVS shunt, and 3.3nF shunt cap.
I'm protecting the digital outputs with a 100 ohm series resistor (1/4W)
I'm protecting the 5V rail by taking a 5.1V Zener and connecting it to the 7805 input through a 470 ohm (1/4W) resistor.  The Zener is hooked up to a PNP transistor that will presumably limit Vbe by pulling the 5V rail toward ground until it shuts off when Vbe goes below a couple hundred mV.

All resistors are automotive qualified, pulse withstanding variety.  I'm a little concerned the 94 ohm series supply resistor may result in too much voltage drop if the power supply drops too low (perhaps when starting).  I'm also a little concerned that if the battery is hooked up opposite polarity the resistors may not be able to handle the heat generated if left connected for a long time as a fully charged battery would cause about 1.56W of dissipation and the max static dissipation is 1.5W (and that's under ideal conditions).  I could throw a schottky in there which would remove the power consumption problem when reverse biased, but would increase drop slightly -- and I'm still left with the problem of handling the energy dissipation from a load dump, which could burn 50+W in the resistors (even if for a short time).  I have no idea how the resistors would take that.

[dom0]

With 7805 types, most of them are of rugged enough design (case of luck or research, I do not know, but it's one reason they have been popular for ages) to be very hard to get to oscillate in any significant way.

The prime reason for this is in my opinion the darlington emitter follower with a very low dynamic resistance at the emitter (rbe in my textbooks). As that resistance is responsible for most part of the open-loop output resistance of the amplifier ... eh... regulator ... phase shift with capacitors at the output, be it smaller values with very low ESR or larger ones with .1 ? ESR, is rather limited. Also the phase margin at "unity gain" of those regulators is rather large by principle, I think over 80 °.

These two points are mainly responsible for amplifier (in-)stability with capacitive loads.