How does noise on a power supply actually affect a circuit?

[TimNJ]

Most of us know that a noisy power supply can cause the circuit its powering to experience problems due to the noise. Noise will appear on the output etc. I know that a power supply with enough ripple to drop under the turn-on voltage of a semiconductor will obviously cause major issues, but noise like smps residue and 60hz and stuff like that.

I'm sure every circuit is different in the way the noise causes problems, but does anyone have an example or two of how the noise actually influences the circuit's operation?

Thanks.

[c4757p]

One quick example - any basic class A common emitter amplifier will directly couple power supply noise to the output. The input voltage determines the collector current into the transistor, and the output voltage is V_CC-(I_CR_C), so whatever is at V_CC goes straight through to the output. This is part of why it's often quite difficult to build a simple audio circuit and keep all the 60 Hz out. (The easy solution is to drop the basic amplifiers and go to an op amp-based solution with negative feedback and something that can properly be called a PSRR, but the old farts frown on that...)

[c4757p]

Another example - a microcontroller running at a low V_DD with its brownout detection enabled could reset if it sees a voltage dip, even if it's not low enough to actually cause faults. This is also one reason why decoupling is so important, because a quick change in current draw can cause a sharp dip in voltage through the inductance of the power traces.

[TimNJ]

Interesting. Yes they both make plenty of sense. Thank you. For something a little more complicated than a transistor like an op-amp, could power supply noise easily get on the output?

[c4757p]

Op amps (decent ones, at least) should list their PSRR (Power Supply Rejection Ratio) in the datasheet. The exact means by which power supply noise will come through to the op amp's output depends on how the specific amplifier is implemented. Lots of (especially older) op amps will actually have the schematic in the datasheet, so that will be a clue.

[TimNJ]

Right-o. Thank you. Your responses are great.

[ptricks]

Micro-controllers are very susceptible to noise in the power supply. Noise on the power supply causes all sorts of weird symptoms like the processor resets, certain ports on the chip work, others don't, values on things like ADC are wrong or unpredictable.

[Razor512]

In my experience, a bad/ noise power supply (noisy power), will reduce your max overclock. The motherboard is usually pretty good at supplying good clean power to the CPU, but not all of the noise gets removed. (even if the noise is still within the ATX standards, having a power supply that does significantly better than the ATX standards, will allow you to get better overclocks)

(the power usually becomes a problem when you are trying to reach the last 100-200MHz, eg pushing your CPU to 4.7GHz VS 4.5GHz)

[c4757p]

(the power usually becomes a problem when you are trying to reach the last 100-200MHz, eg pushing your CPU to 4.7GHz VS 4.5GHz)

At risk of a severe derailment, can you tell the difference??

[kfitch42]

Just saw this over at dangerous prototypes ( http://dangerousprototypes.com/2013/04/29/back-to-the-basics-about-bypass-capacitors/ ):

http://e2e.ti.com/blogs_/b/thesignal/archive/2013/04/23/bypass-capacitors-yes-but-why.aspx?social=social7390694

[Matje]

Interesting. Yes they both make plenty of sense. Thank you. For something a little more complicated than a transistor like an op-amp, could power supply noise easily get on the output?

If you are not careful - very easily ;-)

Imagine an opamp running from a single power rail which should be able to amplify proper AC. Now the trivial way to make that work is to bias the input to half the power supply voltage, by using a simple voltage divider between power and ground (and decoupling input and output with a cap).

Of course by doing that one gets all the power supply noise directly onto the input, just divided by two. Basically all that nice PSRR gets thrown away...

[marshallh]

Put pi filters right before sensitive analog devices/circuits.. i.e. FPGA PLL supplies. Not only does it filter noise coming out from the psu/rest of board but also prevents transients from getting back upstream...

SMPS noise... A board i made long ago had a 1mhz switcher for the core voltage on an FPGA, and being a retard I used alum. electrolytic filter caps. The result was 0.5v ripple on the 1.2v output rail. The fpga still worked fine. Of course I immediately fixed the problem by using the proper MLCC ceramics. Digital stuff is not terribly picky, except for voltage referenced busses.. Noise/SI is more an issue with modern DDR memory technologies where logic transitions are compared against Vdd/2

[ejeffrey]

Op amps (decent ones, at least) should list their PSRR (Power Supply Rejection Ratio) in the datasheet. The exact means by which power supply noise will come through to the op amp's output depends on how the specific amplifier is implemented. Lots of (especially older) op amps will actually have the schematic in the datasheet, so that will be a clue.

Watch out here.  The PSRR listed in the ratings table is usually specified at DC or at <= 120 Hz. If you look in the graph section, you will hopefully find a PSRR vs. frequency plot which will show that it drops at higher frequencies.  For instance, the classic OP27 has a plot showing 120 dB PSRR 1-10 Hz, but falling 20 dB/decade above 10 Hz.  That means if you have a 100 kHz DC-DC converter, you will only see 40 dB of rejection of the fundamental, less for the harmonics.  At 10 MHz, the PSRR is 0 dB.

[AndyC_772]

(the power usually becomes a problem when you are trying to reach the last 100-200MHz, eg pushing your CPU to 4.7GHz VS 4.5GHz)

At risk of a severe derailment, can you tell the difference??

...without using measurement / benchmarking tools, of course.

At these sorts of frequencies, the only useful decoupling / filtering will be that built into the CPU package itself. With decreasing frequency, the motherboard plane capacitance will start to have an effect, then the decoupling caps on the motherboard, then the bulk capacitance in the dc/dc converters that surround the CPU, and last of all, the filtering in the PSU itself.

If PSU noise reduces the noise margin inside the CPU itself, then this illustrates quite effectively just how difficult it is to make a really effective radio frequency filter. IMHO it also shows why running a CPU that far outside its manufacturer's spec is just asking for problems.