Fast switch with picosecond falltime

[uncle_bob]

@Marco
What is the idea with the differential pair current switcher? Please could you elaborate a bit more on your attempt?

Hi

Possibly to be better able to elaborate it properly:

How much of a background do you have in high speed logic / high power microwave / RF pcb layout ?

Bob

[David Hess]

How long does the LED have to be on before it is abruptly shut off?

If an on duration of 10s of nanoseconds is acceptable, then I would try driving the LED with a charge line.  This would work especially well using an avalanche transistor for switching.

What is a "charge line" doing?[/quote]

A charge line is an open transmission line like a coaxial cable which is charged and then a switch discharges it into the device.  In avalanche pulse generators, a charge line can replace the capacitor so instead of producing a narrow fast rise and fall pulse, the output is a wide pulse with fast rise and fall times.  The width of the pulse is twice the length of the charge line.

[dmg]

Laser diodes can, but laser diodes in the blue region are 4 figures.

No, they're actually dirt cheap these days. 405nm, 445nm and 462nm multimode 1W+ CW emmiters can be had for under $100 a pop (even under $50 for 445's) and 100's of mW single mode emmiters can be had for about the same. And I'm talking stuff with a datasheet, sometimes new from factory, sometimes harvested from consumer electronics.

Back in 2011 (or thereabouts) I built a DIY 1W 445nm laser pointer. The diode, a multimode emmiter harvested from a casio projector, was $45.

[Marco]

As I said, blue, not violet. He wanted 480 to 500 nm. I assume he's using something like  Alexa Fluor®488.

That said, I don't really see the need to pulse anything with a laser ... he says the excitation overwhelms the detector, but with laser excitation there is no energy in the fluorescing wavelength at all. Existing machines seem to often use argon ion lasers, so I assume filtering can take care of it.

There's some very cheap argon ion laser modules on ebay from the UK at the moment ... ludicrously cheap. Cheap enough it's hard to resist hitting buy now even though I have no use for them :/ Cheap enough that if I link them I doubt OP would be able to buy them (Of course if they don't work you end up with beryllium oxide you have to get rid off.)

[LaserSteve]

I'm trying to figure out why he wants to do this the expensive, way. I just picked up a used unit for my department.  It uses a Pockels cell to modulate a xenon lamp or an external  laser, adding a little RF in the form of sidebands to the light beam.   But why do you need my 350 picosecond Pockels cell?  Answer is you DONT!

   The Pockels cell has a DC bias and the RF if is summed on top of that to create the modulation.  There is a reference frequency applied to one side of mixer, and there is a signal from the PMT which is the light passing thru the monochromator that picks out the fluorescence. The PMT signal has the sidebands.  A bit of signal processing shows the time shift. A dual lab grade frequency synthesizer that tunes in one hertz steps (PTS D300 , a few are on Fleabay right now) creates the modulation.  A direct conversion mixer handles the signal conversion, and the amplitude and phase shift of this LF product signal shows the fluorescence lifetime.  Calibration is done by placing five out of ten possible cheap known materials in the cell. 

Now a little goggling today shows that my 69,000$ new, now zero dollar donated, unit has been upgraded to use low intensity blue and UV leds to replace the xenon lamp, pockels cell,  and the argon laser.  The LEDs have a DC bias and a RF injection, and now need nowhere near the 10 to 20 watts of RF used to "flip" the Pockels cell.

This can be done cheap...

I quote the description for Review of Scientific Instruments for (Rev. Sci. Instrum. 49, 39 (1978)

Picosecond?resolution fluorescence lifetime measuring system with a cw laser and a radio.


 E. Roland Menzel1 and Zoran D. Popovic1
Quote:
A procedure for measurement of fluorescence lifetimes with picosecond time resolution is described. A cw laser beam is modulated with a standing?wave acousto?optic modulator. The modulated beam is split; one part serves as a reference beam, the other part excites the fluorescent sample. The sample flourescence and the reference beam, attenuated and delayed optically to be equal in amplitude and opposite in phase to the fluorescence, are incident onto a single photomultiplier tube. The thus achieved photodetector ac null is monitored either by an AM radio, whose intermediate?frequency signal is displayed on an oscilloscope, or by a spectrum analyzer. With 30?MHz light modulation and the radio, lifetimes could be determined with resolution better than 15 ps. With the spectrum analyzer and 170?MHz light modulation frequency we have achieved 4?ps lifetime resolution. Correction for photomultiplier transit time versus incident wavelength is made.
End Quote..

Those guys used a DX160 and a fast PMT.. Which if you've ever used a Radio Shack DX160, you'd be amazed that it was that stable...

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

Some Googling:

A portable time-domain LED fluorimeter for nanosecond fluorescence lifetime measurements,
Rev. Sci. Instrum. 85, 055003 (2014)

The above Uses a ring oscillator made of 4049  inverters to drive the led.... It uses the pulse edges to drive a simple transistor circuit using 4 RF transistors  to drive the 405 nm Led,  and a PIC /PIC32MX795F512L-80I/PT for all the signal processing...

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


You can easily inject 30 Mhz RF to a blue LED with a bias Tee...

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

One other thing, if you pulse a blue LED with a nanosecond pulse, the spectrum shifts to the UV, yes,  that's right, fast pulses from avalanche drivers  change carrier lifetimes and cause spectral shifts.  RF modulation causes far, far, less of a shift...

So why try to use a picosecond scope, an expensive modulator or pulsed laser diode, when a blue LED and a simple photo detector  with a lock-in amplifier, microcontroller, superheterodyne receiver,  or DSP will now do this for a fraction of the cost?

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

Marco

BTW< the BeO in the cheap air-cooled argon laser is a tough, solid block or rod of ceramic with an added binding ceramic such as 1%  Alumina. Only dust in a certain micron  size range is potentially super-hazardous, and if your not using a diamond saw or diamond grinder on the BeO, you have no problems...

        I'm a professional laser  user, and we either pay shipping  for the surplus BeO to be recycled by the laser maker, or toss them in normal trash disposal, legally. Usually if a tube cracks we just spray it with a water squirter and bag it, and toss it. In twenty years of use, I've only ever cracked two, and that was by dropping one, the other was a fan interlock failure.   Cracked tubes don't generate enough of the dust particles to make me worry. Solid tubes cleave in BIG pieces with little or no dust.  The stuff is not some horrid alpha emitting isotope causing instant death in two weeks by inhalation. In fact only about 20% of the population is very genetically sensitive and extremely vulnerable to the disease.  You need chronic exposure to really set it off.  If it were such a problem, why does nearly every college and hospital have air cooled lasers venting to room air?

I've been handling BeO for years, and while I respect it, I do not fear using the laser or changing/repumping the tubes.
I do fear the 10 Amp 106 volt,, or 35 amp 250 volts of  DC that drives the arc, depending on the size of the laser.  I rear even more getting an accidental retinal burn...

Kind Regards,

Laser Steve ....

[LaserSteve]

If you insist on pulsing and a fast scope...

http://www.reed.edu/physics/courses/Physics332.s16/pdf/LED%20Luminescence%20Decay.pdf

Steve

[Marco]

or toss them in normal trash disposal, legally.

I wasn't afraid of the material, I simply thought it was more expensive to dispose of than that.

http://www.reed.edu/physics/courses/Physics332.s16/pdf/LED%20Luminescence%20Decay.pdf

"However, it is not a
good LED driver circuit for a waveform capture system. In
our evaluations we used the circuit to drive a 70000 MCD
(millicandela units) blue LED at 2000 Hz. Even with the
room lights off, very little light from the LED was visible to
the eye."

A circuit which drives a LED with logic levels through a 33 Ohm current limiting resistor, compared to one which drives it with 100V through a 50 Ohm current limiting resistor ... they felt the need to compare output power of these by eye?

PS. I like that they build the avalanche pulser on a bog standard breadboard though.

[David Hess]

If you insist on pulsing and a fast scope...

http://www.reed.edu/physics/courses/Physics332.s16/pdf/LED%20Luminescence%20Decay.pdf

Steve

Ha!  Figure 4a shows the charge line suggestion I made although they only used it for testing avalanche behavior in transistors.

They mention the 300 MHz 2440 oscilloscope they used as being too slow but I suspect based on the photograph of their physical implementation using breadboard that they had significant other problems; I do not see even local power supply decoupling.  The recorded pulse they show is not consistent with the performance of a 2440.  At the end they do discuss the shortcoming of their construction method.

Their good results however show that the driving pulse to the LED is not as critical as one might expect and that their construction and performance was completely adequate.

[LaserSteve]

I selected that document because it was not behind a pay-wall.  I work at a academic institution, I have nearly every scientific journal for "free" at my fingertips and can search for ANYTHING... I know the funding agency ultimately pays... But no one bills my cost center for digital library access...

So sending links that cost civilians to have to pay 35$  to see a five page PDF is bad karma from my point of view... The Menzel/ Popovic paper has a complete schematic, and if the OP emailed Menzel I would not be surprised if he was given the source code..

Besides, the art of making a simple  publication useful, is based on looking at the references at the back of the paper..

As for using the breadboard...

Getting rid of the breadboard would help. My jury rigged avalanche circuit hits ~700 pS FWHM into 50 Ohms,  and its just very careful  dead bug on a piece of FR4... No shockline there, either..  For TO-92 packaged devices, the loop area of the circuit is the main determining factor for pulse width..  If I made my circuit SMD, I'd probably break 350 pS...




Steve

[Berni]

Transistors can switch pretty fast in certain modes. I built a Jim Williams pulse generator and got 160ps rise times out of it using a jelly bean transistor from my parts bin. I imagine a special RF transistor is faster. But you do want to keep all connection lengths to a minimum. The inductance of a few cm of wire can screw things up properly.

Also when making such stuff you probably want to have access to a fast oscilloscope with 1GHz of bandwidth. Both to check how fast your light source is switching and to capture the resulting waveform from the detector (That also needs to be really quick)

EDIT:
Oh and on the topic of fast pulses i have a HP8082A Pulse Generator that i measured to have a 800ps rise and fall time. Its an ancient analog dinosaur but is still quite useful today as it can create any form of pulse or square wave signal in differential pair up to 250MHz with up to 5V of amplitude at those rise times. Try asking that from a regular signal generator for under 5 grand.

[Marco]

The pulse doesn't really seem to be the limiting factor in a LED, it's the remaining charge carriers recombining. The ideal drive would be a bipolar pulse, you could probably adapt the avalanche pulser to generate that.

Maybe with some ultra-low capacitance TVS to protect the LED during the reverse current (a 0402 LED can probably handle a 1ns 1A pulse, which won't have a TVS build in).

[uncle_bob]

Hi

So here's a ..errr ... stupid way to do it:

You can fairly easily get buffer chips that are in the < 1 ns rise time range. They will put out < 50 ma, but they are nice and fast. Parallel up 20 or so of them and you have 1 A at a "maybe fast enough" rise time.  It will cost more for your time sorting through chip data sheets than it will for the chips themselves.

Bob

Delay mismatch between channels will screw up every thing. For ns range of rise and fall time, it's okay. For sub ns rise/fall time, may not.

Hi

Give it a try ... it works pretty well.

Bob

[David Hess]

The pulse doesn't really seem to be the limiting factor in a LED, it's the remaining charge carriers recombining. The ideal drive would be a bipolar pulse, you could probably adapt the avalanche pulser to generate that.

Maybe with some ultra-low capacitance TVS to protect the LED during the reverse current (a 0402 LED can probably handle a 1ns 1A pulse, which won't have a TVS build in).

The paper shows how to do that using a unipolar pulse with a parallel inductor.  It mentions empirical testing to get the inductance correct so there is enough reverse bias to sweep the carriers out without destroying the LED.

[Marco]

Sure, but they used relatively wide pulses. I'm not sure if the parallel inductor is the best circuit if you want to excite the LED with really short bipolar pulses. I found a paper which actually gets sub-ns pulses out of standard LEDs (I don't really like their circuit either though).

I suspect that if you turn a LED on for less time you can turn it off faster too, up to the transit time, since the electrons/holes (slow movers) didn't fill the entire depletion zone yet ... so carrier sweep out more quickly empties it.

[Berni]

Sure, but they used relatively wide pulses. I'm not sure if the parallel inductor is the best circuit if you want to excite the LED with really short bipolar pulses. I found a paper which actually gets sub-ns pulses out of standard LEDs (I don't really like their circuit either though).

I suspect that if you turn a LED on for less time you can turn it off faster too, up to the transit time, since the electrons/holes (slow movers) didn't fill the entire depletion zone yet ... so carrier sweep out more quickly empties it.

The circuit they use to reverse the diode for a moment seams like a good idea to me. It provides a predictable reverse pulse so you don't blow your LED due to reverse breakdown.

I don't really get how exactly the actual pulse generator circuit works. But it sounds like one could make that using some fast 74xx logic chips.

Everyone is concerned about the LED, but how fast are light detectors like photodiodes?

[Marco]

I think for fast low noise they generally use avalanche photodiodes, expensive buggers but I assume he has that solved.

How about something like the following? Q1 is set to avalanche on the trigger, using a RF transistor since the LED doesn't really need high voltages (might need to change the trigger circuit a bit for reliable triggering, a bit of negative bias on the base for instance though it would require a large trigger voltage). D2 is a sub-pF TVS. R2/C2 are slightly larger than R1/C1, R3&R4 are significantly larger (they mostly just bring the circuit back to equilibrium). I think this gives you enough free parameters to shape the bipolar pulse, except the rise time of the initial current pulse which is purely down to the avalanche transistor but you want that as fast as possible any way (might want to bypass "part of" R1 with a small cap for some peaking even).

[LaserSteve]

I'm liking the two circuits in the attached pictures

From
An ultraviolet nanosecond light pulse generator using a light emitting diode for test of
photodetectors. Tsutomu Araki, Yasumitsu Fujisawa, and Mamoru Hashimoto


Link: http://scitation.aip.org/content/aip/journal/rsi/68/3/10.1063/1.1147943

At least the ZTX-415 is made of Obtainium and is in production.. 

VA is 520 to 630V according to the publication. It would work at much lower voltages but the higher voltage reduced timing jitter walk. Cmax is 60 pf, above that the led is damaged. Cs in the lower circuit is for impedance matching, value ~ 330 pF,  and R is about 60 Ohms, not 50, for best matching.

The risetime on the shockline circuit was not measured in detail,  but the scope picture is very, very,  nice.... The shockline pulse is flat-topped, the sort of thing you would want for checking scope bandwidth... However, it is a long pulse, having very desirable sharp edges.

 I quote: "Because propagation speed of the current in the coaxial cable is 0.2 m/ns, resultant pulse width W in nS,  is given as W  = 10 D, where D is the cable length in meters"

Sadly they do not mention  the cable type...

From looking at the scope photos, the emission mechanism for the UV shortens the optical pulse considerably, while the visible emission goes long with an afterglow.. So they used a UV filter for the obvious reasons..

Two LED, circuits operating with a slight delay between them, and combined with a beamsplitter, were then used to test the recovery time of their PMTs...

Their work is based on measurements done with a 12 Ghz scope and a 10 Picosecond sampling scope, so this paper is very valid...

LED used is a Model NLPB 520. from Nichia, and the spectrum splits,

I Quote:

"The 380 nm light originates from the InGaN/
AlGaN component, while the 450 nm light comes from the
doped Zn to which the energy is transferred from the InGaN/
AlGaN. Under the conventional dc operation below 30 mA,
the 450 nm light is dominant, and the intensity of the 380 nm
light is not appreciable. When a current larger than 50 mA is
applied, the intensity of the 450 nm light saturates, but that
of 380 nm light continues to increase. The 380 nm light
appears and fades rapidly, while the 450 nm light exists even
after extinction of the 380 nm light when applying a fast pulse"

Not as fast as things could go, but does not need SRDs made of pure  Unicorn horn, the 415 is available from Digikey... I'd do it with logic gate edges, but that's just me...

Curiously, led life was limited to  ~1.0 x 10^9 shots at 20 Khz, not from overcurrent or overheating, but from shock waves to the Die?...

From:

An ultraviolet nanosecond light pulse generator using a light emitting diode for test of
photodetectors. Tsutomu Araki, Yasumitsu Fujisawa, and Mamoru Hashimoto

Review of Scientific Instruments 68, 1365 (1997); doi: 10.1063/1.1147943

Steve

[Marco]

Not as fast as things could go, but does not need SRDs made of pure  Unicorn horn

Jay Diddy avalanched a BR505 in 100 ps, you don't really need SRDs to go sonic fast ... there's the SiGe comparators too of course, but avalanche so easily goes to massive current without multiple stages of amplification that it's hard to resist.

So did they put a divider probe across the diode or use a shunt? I'd not trust that they didn't overcurrent it without seeing a waveform.

[LaserSteve]

They don't say, but if you own both a Hamamatsu Photonics OOS-01,
 and a 12 GHz sampling Iwatsu SAS-8130 with SH-4B plug-in you probably know to use  a shunt.

Steve

[Marco]

They don't say, but if you own both a Hamamatsu Photonics OOS-01,
 and a 12 GHz sampling Iwatsu SAS-8130 with SH-4B plug-in you probably know to use  a shunt.

Maybe, but on the other hand the value of the voltage and current waveform in explaining the workings of their circuit is plain to see too.

Energy wise ~ 1 microjoule of energy, in lets say 500 picoseconds with most of the energy dissipated across the transistors with around 200V drop gives average current of 10A ... hmm, less than I expected with 600V. Guess if they keep the legs on those transistors and the led long enough it's not so bad.