Bandwidth effect on pulse rise time measurement?

BW * Trr = 0.35.

works for gaussian response, which a decent scope should have.

So, 100 MHz == 3.5 ns.

What attenuation does the nominally 50MHz scope give to a 50MHz signal? Manufacturers can be optimistic is it 3dB down or almost nothing at all?

Slightly smoother curves but otherwise indistinguishable.

Even a 1MHz scope would track that slow 15us rise fairly well although it might not be able to show the steepest parts of the curve correctly.

Things get hairy when the bandwidth of the scope is comparable with the rise time of the signal. If you can oversample by 10-20x then the rise time you measure will be dominated by that of the signal itself.

To a first approximation the time axis is blurred by 1/(2f[max]).

How fast a rise time do you need to be able to distinguish? What is the highest frequency waveform you are ever likely to want to capture with the new unit?

Good scopes used to have gaussian responses, BW*Tr=0.35. Lately most everyone is peaking their responses, which causes step overshoot, to claim more bandwidth.

Based on measured Tr, my "200 MHz" DPO2024 is a 180 MHz scope.

Used to be that composite rise time was the sq root of the sum of the squares of scope and signal rise times, so one could math out the scope risetime, within reason. That's not as accurate with a peaked scope response.

You can't, because the information is ambiguous.

What you're seeing on the scope could be the unvarnished, absolute truth of what the signal actually did.

Or it could just be your scope's weak approximation of a pulse that rose "instantaneously".

If you happen to know that the signal did rise "instantaneously", but the scope didn't track it, then you're the wise one; your belief must be based in something, like having the knowledge of the signal's rise time.

But if you don't know the signal's rise time, then you have no basis for challenging the view that the scope is in fact presenting it exactly as is.

:)

And also steps only.

A lot of people over on the EEBlog forum have gotten themselves quite excited at the overly optimistic "rise" times measured with pulses rather than steps (Jim Williams' famous avalanche pulse generator).

We try to correct them, but alas, it's so very hard to convince someone when they read "rise time 1.15ns" on the screen, and The Instrument Must Be Right!...

Tim

One of the sadder days of my professional life was about 10 or 11 years ago when I had to explain to a group of *Tektronix factory engineers* that a scope lives and dies by its step response.

They were trying to sell me this $90k scope that had *seven percent* overshoot on its step response. I bought a refurb version of their previous model (TDS 7704A iirc) that had almost the same BW (7 GHz, 20 Gs/s) but a much much cleaner step response, and was a bit more than half the price. (This was IBM's money and not mine, of course.)

I interviewed at Tek in 1987, back when giants still walked the earth. They wanted me to work on an electro-optical ADC based on lithium niobate Mach-Zehnder interferometers, which was an interesting idea. Although my and my wife's families were and are in the Pacific Southwest (of Canada), I didn't take the job because their lab wasn't nearly as well equipped as the alternatives (IBM Watson and HP Labs on Page Mill Road).

They never did get their revenue per employee high enough to pay competitive salaries for people they really wanted, which was and is a pity. I still read with admiration the works of the Tek guys from their glory days, but those are unfortunately long past. :(

Cheers

Phil Hobbs

Yes, or at least an approximation.

When the signal risetime is less than triple the rated time of the instrument, expect it to be inaccurate.

Roughly speaking, rise times add vectorially: t_meas = sqrt(t_instrument^2 + t_signal^2)

At a ratio of 3:1, you get 9+1 inside the radix, or a 1/10th error. The sqrt of which is about 20% error, so you'd be left wondering just how accurate the measurement is.

a 10MHz system might be expected to have a risetime of 35ns, so your 15us signal would be fine. A 200ns signal would be marginal, and 100ns would be expected to read with some error (on the order of 106ns).

Regarding accuracy, uncertainty due to unknowns in system performance (i.e., have you precisely measured its response?), in the shape of the signal itself (can you resolve all the squiggles and ringing, which will affect the measurement?), and how they interact (does the system bandwidth follow a simple couple-pole Bessel response, or is it high order, or peaked?), mean your effective number of digits is about 1 in this region. So, if you have measurement cursors that say "35.24ns", do feel free to ignore most of those digits as erroneous...

Homework: if you measured a 42ns risetime on a 10MHz scope (assuming the usual rules apply), what's the actual signal expected to be?

Tim

I bet you went to Law school :)

Jamie

We demo'd a R&S scope. I told the guy that it had a lot of overshoot, and he said "everyone is doing that now." He was right.

I sort of always regretted not working for Tek for a while. Now I feel better.

No, the math is pretty simple.

t = sqrt(a^2 - s^2)

where t = true signal risetime

a = apparent rise time, seen on screen

s = scope's rise time

That works if both the scope response and the signal rise are reasonably gaussian, and if you don't try to push it too hard. You can't measure a 1 ns rise time very well with a 5 ns scope.

One can apply a FIR filter to scope waveform data to increase the apparent bandwidth, or to correct for scope/cable defects like overshoot or reflections. That might get you 2:1 bandwidth improvement until the noise explodes in your face. Determining the FIR fiter coefficients is called "the deconvolution problem", a proud member of the family of "ill-posed problems."

No, the math is pretty simple.

t = sqrt(a^2 - s^2)

where t = true signal risetime

a = apparent rise time, seen on screen

s = scope's rise time

That works if both the scope response and the signal rise are reasonably gaussian, and if you don't try to push it too hard. You can't measure a 1 ns rise time very well with a 5 ns scope.

One can apply a FIR filter to scope waveform data to increase the apparent bandwidth, or to correct for scope/cable defects like overshoot or reflections. That might get you 2:1 bandwidth improvement until the noise explodes in your face. Determining the FIR fiter coefficients is called "the deconvolution problem", a proud member of the family of "ill-posed problems."

Aha, so that completes the answer basically. From the scope's bandwidth we get a rise time. Then using this root of difference of squares, we can estimate the actual rise time of the signal.

So it's kind of rubbery. If the scope's rise time is, sy, 5 ns, and the signal also rises in 5 ns, the apparent rise time won't be 5 ns, but something slower. To get the scope to rise in 5 ns, there has to be a step (or at least a rise quite a bit faster than 5 ns). If it is pushed less hard, it complies less.

Only problem is, how accurate is the 0.35 formula for estimating the scope's rise time from MHz rating, and how meaningful and appropriate is the MHz rating.

For this t = sqrt(a^2 - s^2) to work, it seems we need pretty accurate knowledge of s, since it is quite sensitive to the difference between that and a.

Of course, we can obtain it experimentally, by feeding a known clean fast step pulse into the scope (t ~= 0), measuring apparent a, which is ~= s when t ~= 0.

Effectively, this step signal serves as a calibration reference.

Then we can feed in the signal whose step time is unknown and use the formula.

Why bother about doing measurements with electronic doorstops.

w.

Thanks for that formula "BW * Trr = 0.35"

I'm guessing Trr is the 'scope rise time? (Although I'm having trouble understanding how anything other than a *signal* can have a rise time? Mine stays firmly on the bench.)

So it's a parameter of a 'scope that has to be calculated rather than being explicitly specified as part of the manufacturer's spec, yes?

I can't find any mention of 'rise time' in the specs of either my obsolete Pico ADC-200/50 or my likely replacement, the Picoscope 25 MHz 2205.

But from your formula I calculate the latter would have a 'rise time' in microseconds of Trr = (0.35/25) * 10^-6 = 0.014 us = 14 ns

Is that right?

If so, then am I also right that a 25 MHz 'scope displaying a signal's rise time (i.e. the time to reach roughly 70% of its peak) as 1 us, could be confidently taken as very close to that?

--------------------

I think you've provided the particular formula I was seeking later in the thread, in your response to Kaz's post, namely:

t = sqrt(a^2 - s^2)

where t = true signal risetime

a = apparent rise time, seen on screen

s = scope's rise time

Whether the 'scope is 10 / 25 / 50 MHz BW, plugging in the appropriate numbers confirms that all three reporting that signal's rise time of

IOW, my apprehension about 'downgrading' from 50 to 25 mHz is quite unfounded. As confirmed by the first reply I immediately understood, from Phil Allison, and also from other responses after some coffee and further study ;-)

Thanks for all the replies - learnt a lot.

Sounds critical of something or other, but could you clarify please?

I'm betting it's heavy.

One measures scope rise time by applying a very fast step to the input. If the step is way faster than the scope, just read the indicated 10/90 rise and that's the scope risetime.

People used to use tunnel diodes to make a 25 ps step, plenty fast to characterize 30 MHz toob scopes or even low-GHz samplers.

PSPL's first product was a "reference flat pulse generator", used for things like this. Tek recently bought PSPL.

It should be measured, to ensure that the step response is clean.

Old scopes used to do it right.

Yes, if the scope is reasonably gaussian.

Yes.

According to the t = sqrt(a*a - s*s) formula, because 1 us >> 14 ns, we can neglect the s*s term, and go with t = sqrt(a*a) = a, taking t to be 1 us.

What is the error? If we earnestly plug in the 0.014 us for s:

sqrt(1 - .014^2) =~ 0.999902 us.

The error from ignoring the scope's response is only 1 part in 10000!

That is to say, the rise time of the signal is 71X slower than what the scope can do, and so in this situation, the effect of the scope rise time is insignificant; it tracks the signal almost perfectly.

The error is way below the two significant digit precision we have in the 0.014 us figure, so we would be silly not to ignore it. (And I'm assuming that the 1 us apparent rise time given as an example is really

A sine wave that rises from trough to peak in 15 us is oscillating at only 66.6 kHz.

OT: by the way John, where are you getting these bizarre reply subjects from? Reader bug?

Tim

-- Seven Transistor Labs, LLC Electrical Eng>John Lark>

I'm betting it's heavy.

One measures scope rise time by applying a very fast step to the input. If the step is way faster than the scope, just read the indicated 10/90 rise and that's the scope risetime.

People used to use tunnel diodes to make a 25 ps step, plenty fast to characterize 30 MHz toob scopes or even low-GHz samplers.

PSPL's first product was a "reference flat pulse generator", used for things like this. Tek recently bought PSPL.

It should be measured, to ensure that the step response is clean.

Old scopes used to do it right.

Yes, if the scope is reasonably gaussian.

Yes.

--

John Larkin Highland Technology, Inc

lunatic fringe electronics

Beats me. I use Agent and hit "reply." Usenet sometimes does weird things.