let's say I have a pulse with 1V amplitude that last 1msec. and it takes 1ns to rise and fall to that amplitude. and lets say I would like to see it with an ADC, what parameters do I need in order to make sure I don't miss it?
I'm thinking... 1msec pulse.... I need to sample twice the frequency so 2kHz?
is that it?
is that the "bandwidth" I need to look for in my ADC? greater than 2kHz?
also... from what I understand SAR ADC's are faster (larger bandwidth) than sigma-delta's. but for resolving a short pulse like this, with as much resolution as I can get... would a sigma-delta be better?
If it is, why not have the pulse, given that it is quite long in duration, activate it's own sample-and-hold. (I'm assuming you want the amplitude?) ...Jim Thompson
--
| James E.Thompson | mens |
| Analog Innovations | et |
You are not giving enough information about what you want. Jim Thompson has a good solution, if you want the amplitude.
The Nyquist criterion, the 2X frequency, is for pure sine waves. The classic analysis of your pulse (square wave) is via Fourier composition from many sine waves needed to create a sharp rise - Odd harmonic addition for square wave.
So if you want to nail down the exact time onset of the pulse, then a high speed comparator or hi speed ADC might be in order. Rise time directly linked to BW.
s 1ns to rise and fall to that amplitude. and lets say I would like to see it with an ADC, what parameters do I need in order to make sure I don't mis s it?
2kHz?
han sigma-delta's. but for resolving a short pulse like this, with as much resolution as I can get... would a sigma-delta be better?
yes, it's randomly occuring, usuallt the system just has a pretty mellow DC output I'm going to be looking at with an ADC, but every once in a while i t might spit out a spike, I realize 1msec isn't much a of a spike, but I'm just trying to understand what i need to look for in the specs, and what ki nd of ADC would be best... if i can understand this then hopefully i can tu rn 1msec into 1usec and figure it out for that too
I'm making some assumptions here but I would setup an interrupt on the rising edge (not necessarily a software/microcontroller solution, could be all silicon) to sample around the expected middle of the pulse (or 'n' samples and average them). Why bother sampling where the pulse isn't?
The sampling theorem, whether it's attributed to Nyquist, Shannon, Kotelnikov, etc, says that a *signal*, bandwidth-limited to F, can be fully reconstructed from samples taken at 2F. No mention of pure sine waves in there.
Not necessarily. The 1ms pulse on its own isn't a single frequency it is a set of harmonics that sum to make a rectangular shape.
If it is a single pulse in splendid isolation lasting 1mS (or given your stated parameters 1ms - 2ns) then to ensure that you get at least one point on the top of the pulse you need to sample at least that often ~1kHz. Oversampling by a factor of at least 1.5-2 is sensible.
But if it is a stream of pulses 1ms wide then it depends critically on how long or short the time interval between successive pulses can be.
The waveform could in principle be all "pulses" back to back separated only by a 2ns recovery time - taking an extreme limiting case.
You will need to look at dead time correction to determine what your sensor will do if it is triggered too rapidly. (and you would need a ~500MHz sampling to see it reliably)
You can probably find a multi channel analyser off the shelf to do this for you. First question is how fast and how accurate does it really have to be for the pulses that you actually want to measure and classify? There is a related fast ADC thread nearby with some pointers.
You're getting a lot of different answers because your question is vague.
Sampling at 2kHz may be enough to reliably tell that a pulse actually happened, which is what "resolve" sometimes means (although usually one would use "detect").
Sampling at 2kHz will be woefully inadequate for detecting the timing of the pulse to precision greater than 500us, or for reliably telling how wide it is, or for telling the amplitude, etc. I assume that one or more of these is important since you are concerned about accuracy.
As Jim pointed out, if you just want to get peak amplitude or some such, then using the pulse as a trigger may be the way to go.
So it's probably a good idea to tell us what you're really trying to do, and then we can help you out with solutions that will really fit your goal.
--
Tim Wescott
Control system and signal processing consulting
Or for an analog method, a ping-pong gated integrator. Run two integrators of, say, 3.5 ms aperture time, with a 250-Hz rep rate. Phase them so that one is centered on the 500 us gap in the other.
That way any pulse is guaranteed to be completely inside at least one of the integration gates, and you can digitize it at your leisure, more or less.
It's actually a good idea to have all the limit-of-knowledge theorems, but it's VERY hard to apply them. The 'fully reconstructed from samples taken at 2F' comment is wrong, for instance (there are two components of the highest frequency, a sine(2*pi*F*t) and a cosine. You lose one completely if you sample at 2F.
Another complication is Gibbs phenomenon (square waves always get reconstructed with overshoot).
The pulse was characterized by rise time, a duration of constant value, and a fall time. So, if you need to verify those four items, it'll take at least four independent measurements (because there are four independent components of the result, there must be four equations, one for each unknown).
To measure a rise time, typically one would capacitor-couple into an ammeter, and capture current peaks. To measure fall time, capture negative current peaks. And to measure amplitude, you can integrate the signal between the peak-capture time of the rise, and the capture time of the fall. To measure duration, start a ramp at rise-time, and stop it at fall-time, and digitize that value (this is a TAC, time-amplitude converter).
Then you take those four measurements and find the linear combinations that are the actual risetime, falltime, amplitude, and duration. The duration from the TAC includes half the risetime and half the falltime in addition to the duration (and the other values are likewise related in complex fashion to the measurements).
1ns to rise and fall to that amplitude. and lets say I would like to see i t with an ADC, what parameters do I need in order to make sure I don't miss it?
kHz?
an sigma-delta's. but for resolving a short pulse like this, with as much resolution as I can get... would a sigma-delta be better?
Thanks for all the replies,
I'm just interested in reading current, through an ADC, 0-100A, I expect th ings will usually be pretty slow, just a load consuming 2-5A, but if someth ing bad happens, a short or something, I expect a current spike, and I want to see that current spike, looking at "fuses" I see an 8A fuse that pops o pen in 10mS for 100A, so I'd like to mimic something similar to that....
so in general I'm just monitoring and converting slow signals, but I want t o respond to fast current spikes as well
But you are right of course, my statement is a mathematical limit, applicable to a brick-wall bandwidth and an infinite set of samples. In practice, you can only get so close. Let it rest at that.
There are no perfect square waves either. Gibbs phenomenon is no problem for bandwidth-limited 'square' waves, provided you sum up enough sinusoids to satisfy the sampling theorem.
Indeed. But it is lost by the very nature of doing the sampling of a time series. Conversion to a Fourier series is a generalised rotation and requires exactly N values, the DC component and the alternating component both having zero imaginary parts with the other N/2-1 complex components occurring as conjugate pairs in the classic real to complex transform. The limit of knowledge theorem is exactly correct. You can always compute a precise forward and inverse transform on these data.
Back when memory was short and expensive it was convention to pack the N data elements so that the alternating component was stored where the imaginary part of the DC component would be to do in place transforms. It is a convention still followed today in several 2D and 3D FFTs.
Others insist on having a transform length of N+2 and store the zero imaginary components of DC and alternating components explicitly.
The serious practical difficulty is in ensuring that the signal you are digitising is truly bandlimited at F in a way that makes your sampling method valid. It is for this reason that oversampling so that out of band frequencies are all very strongly attenuated is so important in practice. Otherwise they will alias down into the measurement band.
Otherwise you get picket fence Moire pattern style aliasing of higher frequencies reflected around the sampling frequency.
As does any discontinuity in the waveform but that is in the nature of all brick wall band limited filters too.
I think the OP needs to describe what it is he wants to measure. I had assumed something like a photomulitplier tube output where amount of light and duration determines how much oomph was in your neutrino or gamma ray.
Then in general, you will *never* have a square, exactly 1ms wide pulse. How often you need to sample then depends on how fast you want it to respond to an overload condition.
If you want it to act like a fuse, a sample rate of 1kHz would be fine, but that seems like an awful lot of trouble to go to just to end up with no actual electrical protection.
Fuses protect wiring, not components, so your circuit is still just as dead either way. And such a convoluted circuit is never going to pass for UL/NFPA protection so it'll never work as a protection device anyway.
If you want actual electrical protection, use active current limiting (if this involves a switching supply) or an electronic fuse. Mind that said fuse will have to open in maybe 10 microseconds, not 10 miliseconds, to protect itself, let alone whatever's being shorted out. It will never behave in the presence of massive turn-on surges...that's the whole point.
Nope - hasta be a sine wave. Here's why: The Nyquist says you can reconstruct at 2F. A non-sine wave, like a square wave, has higher frequency components. If you could reconstruct other forms of waves, ie non-sine, then you would be re-constructing at
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.