New EE Lab - HF Design

[bcarso]

The FDV301N looks like a decent part---low thresh and lower capacitance than others for the same Ron. I am a little concerned that the speed overall will still be a bit slow especially on turnoff. If you put a 4.7k R from LED cathode to ground you will get the weak (~1mA) "Off" bias suggested, so you aren't working so hard charging the LED capacitance. That will help. But I still think you need some speedup by pulsed current on the leading and trailing edges of the logic pulses. For the 10mA on transition this is fairly easy (a series R-C in parallel with your 200 ohm R) but more difficult for the other transition.

I may do a little modeling---PM me with an email and I will send you a file you can read (and post if you like), if the lab results tonight don't get you where you need to be.

 

[bcarso]

PS: "We should be fine with our input impedence being around 100k, right?"

Your input impedance with the FET is mostly capacitive, and will be ably driven by an HC-series part like the 74HC74. In fact it would probably manage the 10mA of directly driving the LED through a network similar to that shown in the LED datasheet, especially if you use one of the unused FF outputs as opposed to the one that is part of the shift register feedback.

 

[PRR]

> how to keep the transistor from saturating?

At left: no transistor can possibly saturate. And it looks cool in the dark. 27MHz is not an impossible pulse-rate for the faster tubes.

At left: this isn't right, but it is possible to drive a diff-pair so it never saturates. If you also want a small keep-alive trickle in the LED, drop a resistor to ground.

 

[Ian MacGregor]

Ok so here are our lab results:

Using the MOSFET as a switch, we were able to get a max speed of about 3MHz. This was built on a breadboard, using a standard size Red LED to test and an added 4.7k resistor from cathode to ground for our "off bias" current. We drove the gate of the FET with a 0V to 6V signal from a signal generator. I measured the output by placing the scope leads across the source resistor. Another student also mentioned that he had tried a similar configuration and had achieve speeds up to about 8MHz. His was soldered on a prototyping PC board that had a copper ground plane on the top.

Oh yeah, we found out that we are expected to have a max switching speed somewhere beetween 10-20MHz.

We did not try any RC combinations to speed up the circuit as the shop was closed and we couldn't get any parts. I understand the concept of using a RC circuit to speed up the switch time, but I am a little unsure on how to compute values for R and C. Do you guys think that we should abandon the single MOSFET switching stage? The rest of the students are using the buffered differential circuit given by the professor. Most of them have not been successful.

Thanks for the help!

Ian

Super Late Night Edit:

Ok, I've been thinking about the RC spped up network a bit. First, here are the values I've been using for various parameters:

Cdiode = 26.5pF
Idiode = 20mA
Vf = 2V

Rsouce = 180~200 ohms

Ok, so I started re-reading PRR's post about Slew rate and figured out that the slew rate for our diode/voltage comination is 20mA/26.5pF or 0.755V/nS.

This is where I think I might be taking "mathematical/physical leaps":

If we divide our slew rate by the voltage we want to see, we get the risetime: 2V/0.775 = 2.65nS. If this is the 10-90% risetime, then it corresponds to a time constant of 1.2nS. If we use a series R-C speed up network in parallel with our source resistor, then (assuming the R to be 330 ohms) C = 3.6pF. Does this make any sense at all?? I have a feeling that I have completely lost it! :shock:

ehh...

 

[Svart]

The FET i suggested was just a suggestion, I'm sure you could probably find a much better one. I use this one: http://www.irf.com/product-info/datasheets/data/irlml2803.pdf for various things, it might be even more suitable for you due to the lower RDSon however the gate threshold voltage is a bit higher which might tend to keep the FET in partial conduction unless your gate voltage is a stiff 5-6v.

there are so many limiting factors to switching at super high speed like this. as for your classmate who is switching at 8 mhz, how does his design differ from yours besides the groundplane? is the gatedrive signal going through a speed up arrangement or being driven directly from logic, or is it being driven by a FET driver?

have you scoped out the switching and watched the actual waverform to see if it's being limited by risetime or falltimes?

I would expect a "normal" led to be much slower than the LED designed for this kind of speed. you might indeed be limited by it.

as for RC networks, what is the proposed circuit? can you post a schematic?

:thumb:

EDIT: did the professor give you the impression that it was entirely possible to attain speeds like this with a single switching element, or do you think the lessen is not the top speed but to learn how things interact and how to make them work for you and thus the 27mhz is just a goal number? just a thought. :green:

EDIT2: I just had another look at the LED specsheet, it says the Tr and Tf are ~8nS which is within most high speed FET's ability. the ON and OFF times will be logic controlled and the switching transition times are limited by the above mentioned situations.

also..and i don't mean to keep rambling, but this stuff is interesting to me, would/could you use a totempole type switching arrangement that will pull the diode up/down in overlapping waveforms faster than a single switch could do by itself?

and again, keep the drive IC CLOSE to the FET.

 

[Ian MacGregor]

I'll try and post an updated schematic in a sec. The guy that got 8MHz, his circuit was the same as mine posted above. The only difference was that his was soldered, it had a ground plane AND he was using the actual LED we must use. His gate was being driven directly by the logic, I think. When we were testing our circuit, I did take the LED out and replace it with a resistor that dropped about the same voltage. Our switching waveform looked the same and performance was not any better. This led me to believe that the LED was not our limiting factor.

The professor did not tell us that it is (or isn't) possible to achieve these speeds with a single switching element. I just wanted to try something simple before diving into a multi-transistor circuit without really knowing what is going on. Also, I am learning a LOT about the limitions of devices. If I just built the recommended circuit blindly and it worked, I probably wouldn't have learned as much.

The 27MHz goal is just set by us because that is how fast we got our random generator to work. To pass the lab, we should be able to switch at least at 10MHz.

also..and i don't mean to keep rambling, but this stuff is interesting to me, would/could you use a totempole type switching arrangement that will pull the diode up/down in overlapping waveforms faster than a single switch could do by itself?

This may be what we need to do now. I am about to leave to go discuss the lab with our professor at his office hours. Hopefully all the students aren't there. I'm not familiar with the totempole arrangement. Is that like a differential stage?

Ian

Edit: Here is the updated schematic: http://groupdiy.twin-x.com/albums/userpics/10021/FET_Switch-2.pdf

We haven't tried the RC speed up network yet.

 

[bcarso]

I wish there was a decent model for the LED. I found that the red LED model in circuitmaker is horribly slow and very unlikely anywhere near the part on the assignment. I am surprised that your results didn't show more of a change when you changed LED's.

Again, we know that if we trust the datasheet for that part they get the performance displayed using the basically voltage source drive with resistors and speedup capacitor. The current is not as precise as it could be since it depends on the forward drop of the LED which will be variable from part to part and temperature sensitive.

High speed CMOS is pretty darn fast and hard to beat, and has adequate current for this application.

 

[Svart]

http://groupdiy.twin-x.com/albums/userpics/10015/FET.pdf

How about give this one a test?

I *think* you might be better off with the current limit resistor in the source to the led. I might be wrong though..

Soldering and proper PCB routing is crucial in such high speeds. If you look inside any highspeed RF type device the groundplane will cover both the top and bottom, usually made of something highly conductive like gold(for speed too..) and usually shielded beyond anything we see in daily DIY or work. You'll also start to see strange things like curved traces instead of right angles since electrons tend to build up in corners and other screwy things like that at high speeds..

you might try an etched PCB with the shortest traces you can handle, make it SMD and put ground plane all around the traces.

there are plenty of RF transistors specially made for switching into GHZ range, although they might be expensive.

sounds to me like the professor really wants for you to learn the limitations of the parts rather than just building a winning design..

A totempole is similar to push-pull, here is a simple totempole i drew up for you:
http://groupdiy.twin-x.com/albums/userpics/10015/TPOLE.pdf

as the gate signal nears 0VDC, the Pchannel turns on and pulls the output up, and inversely as the signal goes up the Pchannel stops conducting and the Nchannel start pulling down. usually you have a few more parts to keep the overlapping( also called shoot through..) to a minimum but that is the basic idea.

:thumb:

 

[bcarso]

PS: try driving the thing from the Q bar output of the generator with the R-C arrangement in the spec sheet. I'll bet it works fine. Use Q bar because it is an independently buffered output that isn't being used for the XOR input.

(Added after some more thought): I also notice that the manufacturer's circuit is hitting the part with 30mA, which is a bit close to the abs max particularly if you are going to have it be even higher running your logic from 6V. So I guess at least scale the resistors up to get back to <30mA, and reduce the cap size proportionally.

 

[Svart]

yep, as I read what Bcarso has mentioned, he is indeed correct, you could likely just drive the whole thing from the logic like the specsheet and leave out unnecessary parts..

 

[bcarso]

But Svart, I'm having "old fuddy duddy" remorse---I was having fun playing with different ways of doing this!

There are undoubtedly ways to improve on the mfg.'s circuit but Ian and his cohorts are pressed for time.

 

[Svart]

:green: yeah sometimes fun has more to do with design than logical thinking does! and sometimes it even leads to more interesting results!

There are probably 100 different ways to getting the same end result, but for me the fun is actually trying all the different ways and seeing what happens and why. that is the lesson the prof. is trying to teach here I think. sometimes you run into issues that all the simulation in the world can't forsee and sometimes the answer can stare you in the face without being seen.

We'll see what his next tests say and then we'll know better.

As for the time issue, there never seems to be enough of it during the design stage does there?

:guinness: :thumb:

 

[Ian MacGregor]

As for the time issue, there never seems to be enough of it during the design stage does there?

Haha!! Exactly! We talked to the professor and got some very interesting info. I have to breadboard up a new circuit real quick before work, but I promise I will let you know what happens.


Ian

 

[bcarso]

One additional thought: realize that what you want to know is the photons vs time, which is not known until you have a fast photodetector coupled to the system. Looking at the voltage across the part can be misleading when you are trying to go really fast.

When you do set up to look at the effect of the photons, beware of the electrical emissions from the assembly affecting the supposed photocurrent! A ~short fiber link would be advised, and judicious shielding.

 

[PRR]

> impression that it was entirely possible to attain speeds like this...

The GigaBit connection into the new building is optic, basically the same blinky lights that Ian is messing with.

> with a single switching element

No, probably a more snazzy driver, and a snazzier LED or laser. Still, a dozen MHz "should" not be any great problem, after Ian has worked for many years and many designs. But the first design has a high step.

> drove the gate of the FET with a 0V to 6V signal ... Vf = 2V

So your "amplifier" isn't even amplifying voltage: it works at a significant voltage loss. You better have some current gain. IF you need it: have you characterized your logic generator and know how hard you can suck on it? As bcarso says: some of them CMOS chips can dump significant current on their own. I dunno these super-fast CMOS, but plain old CMOS could have several gates paralleled for increased drive, and the cost of an extra 4-gate chip was less than stocking and stuffing a BJT/FET just for one heavy load.

> Cdiode = 26.5pF
Idiode = 20mA
Vf = 2V
Rsouce = 180~200 ohms

You really getting 2V drop in the LED? How do you square this with 20mA 200Ω?

> risetime: 2V/0.775 = 2.65nS. ...time constant of 1.2nS. If we use a series R-C speed up network in parallel with our source resistor, then (assuming the R to be 330 ohms) C = 3.6pF.

I dunno. My math is bad, but "seems to me" the RC product should be similar to the RC in the troublesome load. 26pFd and 200Ω, and you are using a source resistor "about twice" the stated LED resistance, the source cap would be 26/2 or like 15pFd, not 4pFd. Or at least: if I'm wrong, 15pFd should make a for-sure overshoot, clearly excessive on the scope.

> We did not try any RC combinations to speed up the circuit as the shop was closed and we couldn't get any parts.

Greenhorn. Use a gimmick. Twist some insulated wire together, you have a few-pFd cap. In fact, most cable today has pFd specs so you don't even have to guess. Yank some network cable from the wall and look up the specs for CAT5. I bet it is around 25pFd per twisted-pair per foot, or 2pFd per inch. A 4pFd is trivial. A 15pFd is 7": at some point you have to figure speed-o-light and realize the blip won't fill the whole length, but 7" is small at 10mHz, and you could parallel three 2" lengths to go faster.

 

[Svart]

"Greenhorn. Use a gimmick."

you old timers crack me up!! :green:

 

[pmroz]

Twist some insulated wire together, you have a few-pFd cap

Saw that one in Anderton's projects book (spluffer), regular 20-22awg copper wire. Did they not make 5pf caps back then?

_________

 

[bcarso]

"some of them CMOS chips can dump significant current on their own. "

As a current source/sink at 6V the HC gates usually pull 30-40mA in my experience. When they bottom out to either rail they look like about 30 ohms to the particular rail. And of course as PRR points out paralleling is generally easy to do at least with gates in the same package.

The HC and other parts are frisky enough that you can burn them up if you have a floating input and get P's and N's in series on continuously. Hard to do this with the "high-voltage" ~4000 series parts.

"you old timers crack me up!! " Yes by crackey... we crack ourselves up too. Reminds me of that Frasier episode with the curly-haired kid that bedevils Frasier and Niles, but whose girlfriend ends up knowing the wine connoisseur with the legendary cellar.

 

[bcarso]

"Saw that one in Anderton's projects book (spluffer), regular 20-22awg copper wire. Did they not make 5pf caps back then? "

Do you mean 5 micromicrofarads?

 

[Svart]

i guess i'm just a spoiled brat. I have a large set of drawers(old blueprint drawers) with all kinds of parts in them. I just buy a bunch extra of everything I use and put it in the drawer. I use a lot of small caps in the PF range so I have a fairly wide selection to choose from. I wouldn't think twice about ordering a range of values from Digikey or someone else for things like this, but then again I like the resourcefulness of making a couple PF cap from some wires.. :green:

I guess times are changing and both small and large values are easier to come by and cheaper these days.

I'd like to know how Ian is doing with this. :thumb: