broad band noise source vs sweep signal for spectral analysis

[sarepairman2]

Visually the effect looks the same, but are there any subtle differences with the results found using each method? Does a semiconductor (or tube) behave differently when its bombarded by everything at once? how about sweep speed (i.e. some kind of "hysterisis" effect)?

the two scenarios would be

1) amplifier 1Khz-10MHz
2) amplifier 1Khz-40GHz

I guess it has to do with power, because the noise PSD is distributed.. but its interesting to think about.. its hard to make a comparison because at the same power level the noise might be considered "small signal" while the sweep might be "large signal" effects.. required to keep the devices at the same temperature *unless temperature control is used

[Earendil]

To best of my knowledge linear networks don't have this effect or such an effect is insignificant compared to normal VNA measurement uncertainties.

For non-linear networks such an effect is trivially possible. E.g.: I could create a microprocessor controlled receiver/transmitter that behaves differently depending on the sweep speed.

So I think you need to be more precise in your question. I mean to answer your question one have know what kind of circuit we're talking about. A transistor? An amplifier? What topology? Etc...

[T3sl4co1l]

Comparing EMC to shock and vibe (the mechanical equivalent) is interesting.

Normally, EMC does:
- Susceptibility sweep (single frequency with modulation, applied to cables, and with antennas, as appropriate)
- Surges (on applicable power input or load connections)
- Spikes (on cables and pins)
- ESD (anywhere it can touch)
And also checking emissions.

Nothing is measured during the susceptibility tests, and it is assumed that the instruments are calibrated to deliver their specified waveforms into a suitable dummy load (or in the case of the susceptibility sweep, it's usually measured at the final amplifier, but not at the coupling network to, or in the space around, the EUT).  In general, one wouldn't expect significant nonlinearity to occur, so the frequencies the EUT is subjected to will be largely what's given.

Shock and vibe does fully body acceleration, in three axes each:
- Sine wave sweep (low level, no modulation -- somewhat akin to an EMC emissions sweep, even though this uses an external source)
- Sine wave sweep (at specified level -- stress test)
- Surges (usually X acceleration for Y ms, N cycles)
- Random noise (some RMS acceleration, for specified duration)

Acceleration sensors are attached to the shake table and EUT; acceleration is calibrated at the table, but measured in as many locations as desired.

What you see on the readout during the high level sweep (and in the averaged FT view of the random test) is similar to the peaks and valleys of the low level sweep, but depending on construction, those features can be expected to appear flatter (lower peaks, higher valleys), or shifted somewhat higher in frequency, due to the harmonic generation and frequency mixing action (and yes, even parametric oscillators) of wobbly materials shaking and sliding and slapping around.

What you see on the transient readout, is similar to the input pulses, but with much more 'crashing' behavior -- high frequency content that wasn't there before, and which dies down in an inconsistent fashion, as energy sloshes between different axes (only one of which is being measured) and between different degrees of freedom (like how a square plate has spacially orthogonal, but spectrally identical, resonant modes; which are normally independent, but can be coupled together by a nonuniformity, for example a screw near the corner).

So, as poor a grasp as us EEs have on modeling our circuits, my heart goes out to those few MEs who have to delve into the nitty gritty, multidimensional, manifestly coupled, highly nonlinear and inconsistent systems they have to work with.  Which fortunately isn't all that many, but when every gram counts -- aerospace for example -- you have to take the time to optimize resonant modes to require the least amount of damping and such.

Tim

[T3sl4co1l]

As for reasons, there are a few:
- If there's a simple problem (like common mode noise blowing out logic thresholds or coupling in noise), it will be blindingly obvious: the comm channel drops out, or spews gibberish; analog inputs read the modulation tone instead of the desired signal; etc.
- If there isn't a problem (e.g., a robust digital radio code, combined with a high dynamic range receiver, and rapid error detection and correction), then it seems unlikely that other signals (coherent or otherwise) would also break it.  There might always a magic signal that will break it, but the complexity of that signal might be entirely too coincidental (i.e., it's highly unlikely that random noise would look exactly like a given CDMA key, or sufficiently like it to degrade SNR worse than usual).
- And most of all, the nature of interference itself.  There are extremely few sources of very intense, wideband white noise, or sufficiently complex environments where, say, randomly modulated narrowband sources approximate the same thing.  In a 1MHz bandwidth (at 50 ohms, so this is what a receiver would see from a matched, lossless antenna pointing at a black body of the same temperature), you'd get the same amplitude (say, 1V RMS) from an impossible temperature of 3.6 x 10^14 K.  In comparison, the Sun's corona is a few million K, some hundred million times smaller, which should give a signal in the mV.

(Not to say there aren't such sources at all -- Sutro tower outside San Francisco is notorious for beaming out... pretty much everything.  Up close, a modern DTV broadcast -- 6MHz wide, around 5kW -- would look blazingly hot indeed within its bandwidth!)

Whereas the sources of mechanical vibration (which is the first noise test that came to mind) tend to be very nonuniform and intense in nature: road noise, fluid turbulence, the variable frequency whir (sweep + harmonics + noise?) of a motor, or engine and drivetrain, etc.

Tim

[Kleinstein]

For a linear system the slow sweep and white noise response schould give the same result.

There are a few differences if the system is not linear: than different parts of the noise can interact despite of the usually low level. From the practical side the noise source has to be well reproducable which is not that easy or perfect. Here simulaneous recording of input and output side can help.

The sweep needs to be slow, so the system has time to reach a stationary state - otherwise corrections are needed. This is usually not a problem with amplifiers or normal low Q filters. However this can be a problem with high Q filters (e.g. crystal based) especially at lower frequencies. If the scan is to fast the resoance curve can be assymetric. There is also the chance that large amplitudes are involved so that nonlinear effects can become visible - with an amplitude dependent resonace this can look like a hysteresis effect giving rather different curves when sweeping up or down and a sudden drop or increase in amplitude.

[sarepairman2]

I'm sure there are electrical equivalents of all the mechanical things mentioned by teslacoil if you have precise enough equipment to detect it.

and i was more imagining simulating this problem using specifically designed equipment for a expensive physics experiment rather then trying to find it as a (un)naturally occurring EMC problem, but it is all very interesting.

Its interesting to think about how the mechanical, electrical and thermal aspects all tie into each other on a device level.

[IconicPCB]

the answer to Your question will depend on the type of measurement set up You have

Say a spectrum analyser and tracking generator are used then the distortion and dynamic range should be readily visible.

Consider the same specan in conjunction with a white noise source . The envelope of the  transfer function only will be discernible.

Consider a vector voltmeter and a sig geny and the presence of distortion products will become invisible dut to sampling nature of the vector voltmeter.

Consider the same voltmeter and a noise source.. useless.

Consider a Swept frequency set upwith a broadband detector.. the measurement is likely to be affected by off frequency distortion/gain/mixing products.

Same setup driven by a noise source will make absolutely no sense.

So it is not the DUT You need to worry . It is the test method which must be foremost.

One quirk when using Specan and noise source... make shire Your sweep time and IF bandwidth are matched otherwise the transient response of the analyser IF strip will distort the measurement.