I have a project for work I am mulling over. At first it seemed easy, until I found out some specifics.

I need to detect when a 0.2ns pulse exceeds a specific level, around 4 or 5 volts. I also need to make a simple calibrator box capable of generating similar style pulses to make sure the first box is working.

This is a simple Go/No Go project, if it does not see the pulse the light stays green. If it sees something the light shows red. So a simple latch and hold is in order.

My current thought is ECL logic, but I am not sold on the idea. Anything that works is good enough. Basically I will have SMT components on a small PC board hanging directly off the BNC connector (required).

Right now we are using a really nice LCD HP oscope to detect these pulses. They are expensive, and not operator (as in non technical personnel) friendly. The box would signal when it is time to get maintenance/engineering involved.

Other design considerations were easy, and I have basically drawn them up. A 10 second on time from a simple push button, 9V battery (if needed we can use other batteries), low battery voltage indicator. I have finished drawing this part, it was a no brainer. I have also incorporated a variable power supply voltage since this project will require a basic magnitude detection, and I haven't gotten a handle what the final circuitry will look like.

I will post those schematics up later.

 

ECL logic is almost obsolete and may not even be available. I've not seen any surface mount ECL chips. Next, as you know it operates from GND and a negative supply. I have heard of PECL (Positive ECL) where you have a Positive supply and GND but I have limited actual experience with this.

I don't know if discrete logic has kept up with the advance of FPGA and ASIC technology but I'm guessing a mostly empty small geometry CMOS FPGA might be what you need especially if the core runs off of 1.8 V with a 5 V tolerant Pad Ring.

The CLK input of an optimized JKFF macro might just do the job of detecting and latching such a narrow pulse.

The other thing I would do is have the incoming narrow pulse STOP (or inhibit) a free running oscillator. This way you can be sure that if you go from something (a pulse train) to nothing (no pulse train), and can RESET the device to bring the pulse train back that there was an actual event.

For the pulse generation you may need to build something out of discrete parts. Fastest I've ever done with logic was a couple of nanoseconds, so I'm not going to be much help to you there, and that is off by an order of magnitude.

Just my $0.02

 

Hi Bill,

An RS485 differential receiver (SN75LBC179) , or more modern/accurate LVDS version (SN55LVDS33-SP) may work for this application if one of the inputs is biased to the detection level reference say... 4V. You will still need some fast latches following to remember the trigger event and integrate/count... whatever.

Are the pulses repetitive? What period in between?

Regards,
Ifixit

 

Best answer is the pulses are sort of repetitive. They are generated by a mechanical source using something in the neighborhood of 2KV. Think ESD, and you won't be far off. The 2KV can not be eliminated, only the residual drained, which is what they do. If something goes wrong however, we can generate a lot of very expensive scrap fast (ESD again).

 

What are the peak voltages (positive and negative) that the detection circuit will be exposed to?
Are you saying it's 2kV?

 

No, but it could go up as high as 10 V (absolute worst case). There is a usual 3 V pulse there, if it gets too large we need to shut the machine down. Hence the monitoring.

It is a residual voltage left over from the 2KV, caused by stray capacitive my best guess.

I thought this was going to be easy until I realized the narrowness of the pulses.

 

Have you considered that, with 100ps rise and fall times, even 0.5" long PCB traces and wires will act as transmission lines? If they are not source and/or load terminated to the approximate impedance of the trace, the result will be ringing, amplitude distortion, etc.
Your circuit design and layout will have to take this into account.

Do you do LTspice sims?

 

Nope, and they are running this through a standard 3' coax terminating in a BNC connector. This I have no control of, the scope is set for 10MΩ input. Like I said, on the part I do control I will use SMT with the shortest lengths I can. I'm just looking for ideas on logic (or analog) that will be fast enough. It is doable, else how could the oscope see it, and measure the voltage besides? What I want is much, much simpler. Doesn't make it easy though.

I suspect I will have several tests before I even get close.

 

Would this part help? Use the PECL version so it's easier to interface to other stuff.

http://datasheets.maximintegrated.com/en/ds/MAX9600-MAX9602.pdf

 

Here is another comparator: 100ps minimum pulse width. http://www.analog.com/static/imported-files/data_sheets/ADCMP580_581_582.pdf

Input range is quite limited though. And I also would be worried about trace length...

 

Would this part help? Use the PECL version so it's easier to interface to other stuff.

http://datasheets.maximintegrated.com/en/ds/MAX9600-MAX9602.pdf

The original requirement was for 200 ps pulse width. At 250 ps minimum I don't think this part is in the running unless the original requirement can be eased. Ron's point is an excellent one and you do realize that 3' of coax is about a 3 nanosecond delay (@ 1' per nanosecond). I also worry about ringing and distortion. The 10 MOhm impedance of the scope probe is nearly irrelevant; what is relevant is the capacitance which will be in parallel with the distributed capacitance of the coax and any trace capacitance on the PCB. I would not discount the need for some heavy duty RF CAD software, at like 10K per seat, to get this right. LTSPICE is a good start if you get all the modeling parameters correct.

I would focus on having the leading edge of the pulse drive a device to saturation and have the device take a while, relatively speaking to come out of saturation. This would be a "natural way of stretching the pulse to a more manageable width. I still don't have way to generate the little buggers.

 

Linear Technology AN47 - High Speed Amplifiers and digital app note has EVERYTHING you may need for this project.

The pulse generator on page 93 (PDF Page #) is a simple affair, sort of. A transistor forced into avalanche creating a 10V pulse out fairly often.

Pulse is ≈350ps rise/fall time. No idea how to get faster than that.

That app note covers probes to bypass capacitors to construction techniques and has some example high speed circuits.

However, the 200ps pulse you are trying to "trigger" on equates to a 5 GHz bandwidth minimum!

The 74ABT series of Logic may come close to what you need. 2nS/V for inverter Guess not.

 

Searching through videos, I'm wondering if you can Replicate this for the detector, essentially, an LC tank circuit, made out of essentially the coax self inducatance/capacitance in order to extend the pulse into an oscillation. It is started from a nearby static pulse. Don't stop there, boost the oscillation, sample/hold/measure?

Vid 1 on pulser from app note above, measured with a 13Ghz scope

The problem with the detector is building the front end with that sort of analog bandwidth. Half that, and digtal, not easy, but not unobtanium (in low cost, anyway).

Need more details about distances and what's in place currently, there may be a point to detect the static build up prior to it creating the discharge pulse?

--ETA: Are you certain there are no pulses of shorter duration that the scope isn't catching?

 

Can you just detect the average energy of the pulses? That would make life a lot easier. You could AC couple, and then RC integrate using some SMD parts capable of RF type frequencies.

That would turn your 2nS 5v pulse into a 20nS 0.5v pulse which you could handle easily (and more accurately) with an opamp followed by some type of level detector.

 

Can you just detect the average energy of the pulses? That would make life a lot easier. You could AC couple, and then RC integrate using some SMD parts capable of RF type frequencies.

That would turn your 2nS 5v pulse into a 20nS 0.5v pulse which you could handle easily (and more accurately) with an opamp followed by some type of level detector.

It was 0.2 ns or 200 picoseconds. A small detail, but kinda critical.

 

Guys, I apologize. I've been stupid (again). Turns out the oscope in question is a Fluke (number unknown) and it is a 50Mhz scope. So that is what I will design for.

Sheesh, talk about over complicating a problem!

 

Orders of magnitude are everything. At least it wasn't English units vs. Metric units -- LOL

 

I don't think I misread the units, but the bandwidth has to count for something. Dern if I know how they can show such tiny widths on a 10ns scale for a 50Mhz scope. Going with bandwidth however, I should be able to detect the pulses a lot easier.

 

I don't think I misread the units, but the bandwidth has to count for something. Dern if I know how they can show such tiny widths on a 10ns scale for a 50Mhz scope. Going with bandwidth however, I should be able to detect the pulses a lot easier.

So how wide are the pulses? I'll guarantee you won't see 200pS pulses on a 50MHz scope.

 

I don't think I misread the units, but the bandwidth has to count for something. Dern if I know how they can show such tiny widths on a 10ns scale for a 50Mhz scope. Going with bandwidth however, I should be able to detect the pulses a lot easier.

I know it wasn't a misreading of units, I was referring to the Mars Lander that crashed because of English/Metric confusion.