3.3V to 24V 20MHz signal level shifting

[ali_asadzadeh]

Hi,
I have designed this circuit, but It's not working as expected! Do you have a better Idea of how it could be done, for lower frequencies under 1MHz it seems it works somehow, although the duty cycle is changed!
The simulation results for 600KHz and 20MHz are attached!

[Benta]

You have heavy storage problems in the lower transistor, caused by saturation.
You can try to improve the switching performance, for example by using a Baker clamp. And you probably need much faster transistors that the ones you've chosen.
https://en.wikipedia.org/wiki/Baker_clamp

[Zero999]

An AC bypass capacitor across R3 would also help.

And why the heck are you using a 50W zener diode for the current source?
http://pdf.datasheetcatalog.com/datasheet/microsemi/SA5-15.pdf

[Mechatrommer]

and why he even need T1 and R1? i guess the 1N2804 is used because thats the default zener given by the SW when we placed the component into the circuit.

[Zero999]

and why he even need T1 and R1? i guess the 1N2804 is used because thats the default zener given by the SW when we placed the component into the circuit.

R1, T1 and Z1 form a constant current source.

[Marco]

There are a couple gate driver ICs with sufficient voltage range and speed which could do this. IX 4426/27/28 for instance. You get rail to rail switching at relatively low static power consumption, this would be hard to replicate discretely.

[Mechatrommer]

and why he even need T1 and R1? i guess the 1N2804 is used because thats the default zener given by the SW when we placed the component into the circuit.

R1, T1 and Z1 form a constant current source.

right. but does it help in this application? because i tried simulating with just a resistor (VM2), not much worse than with current source (VM3) see attached. that is at 700KHz before both degrading in magnitude by increasing frequency...

[Benta]

Using a current source for the upper leg is good practice and takes some variables out of the circuit. Instead of the Zener, I'd just use two 1N4148s in series, much simpler.
R1 should be significantly larger, say, 22 kohms, giving 1 mA through the diodes.
Not knowing what this circuit is driving, the current through T1, and thus the value of R2 is open, but 100...470 ohms is probably ballpark.

The major problem is T2. Saturated switching at 20 MHz is a challenge, which is why I proposed a Baker clamp (preferably a single Schottky diode base-collector) for T2 to keep it in the linear operating region. A faster transistor is probably still needed, though.

Edit: I just put on my thinking hat. You'll need a Baker clamp for T1 as well. When T2 is off, T1 will also saturate. So two small-signal Schottky diodes need to be added.

[ali_asadzadeh]

Thanks for your hints, I have replaced the transistor with a faster version 2n3904 and got better results, Also I have added the shotckeys, but it seems that the waveform degraded more!

[Benta]

Remove SD2, you only need one Schottky. Place it on T1 instead.

[Benta]

And don't use power Schottkys, they have a lot of capacitance. A 1 A rectifier is completely wrong for this application. Something like a BAT46 would be right.

[Zero999]

I think the circuit is far too basic to drive any kind of load.

Does it need to work down to DC?

Here's a circuit, which the simulator says is capable of driving a 100R load and will work from DC up to about 30MHz. The source impedance needs to be below 200R for it to work properly. It might not perform quite as well in real life. Build it with SMT components, on a small PCB.

[ali_asadzadeh]

Dear Hero999 thumbs up Thanks a lot, would you please attach the simulation results or the File for simulation? is it in Tina or Lt-spice? regarding this high  part quantities, I think the MOSFET drivers that were suggested was the best price, size option maybe

[Benta]

Recognize something on the output transistors?

Yep, BAT54 Schottky Baker clamps.

[Zero999]

Dear Hero999 thumbs up Thanks a lot, would you please attach the simulation results or the File for simulation? is it in Tina or Lt-spice? regarding this high  part quantities, I think the MOSFET drivers that were suggested was the best price, size option maybe

I created it in LTSpice. If you download and install it, you can use it to open the .asc file attached to my previous post and simulate it. Here's a picture of the output, when it's driven from a 20MHz 3.3V square wave and it's been running long enough, to stabilise.


Notice that the component idents go from right to left, because I designed it output to input?

Q1 & Q2 are the output driving transistors, configured as common emitter switches, with the D1 & D2 as Baker clamps. R1 and R2 from a 1.65V reference. If the input signal exceeds this by 0.6V, the output goes low, and if the input goes below it, the output goes high. Q3 and Q4 are connected in common base configuration and level shift the input signal to drive the output transistors. They are only used at low frequencies, at higher frequencies, the signal bypasses them, via C2 and C3. Q5 and Q6 are in common collector configuration and buffer the input signal to provide a low impedance to drive the rest of the circuit. R6, R7 and D3, D4 bias their bases so they turn on slightly, otherwise a  far greater voltage swing, than +/-0.6V would be required to drive the circuit. The biasing circuit only works at low frequencies and is bypassed by C5 and C8 at high frequencies.

I agree, a MOSFET driver is certainly the better solution here, probably not the lowest cost component wise, but the reduced PCB area probably makes up for it.

[David Hess]

Besides baker clamping, capacitively bypassing the base drive to T1 will help and using capacitive coupling from the drive signal to T2 (most recent schematic) will help move charge around for fast switching.

With 2N3904s and 2N3906s, I made a similar current source and sink based RS-232 level shifter a couple years ago and got perfect 50 nanosecond edges but did not check storage time.  20 MHz is a period of 50 nanoseconds and a width of 25 nanoseconds so this circuit is not going to be fast enough.  What you need looks more like the z-axis amplifier in a CRT oscilloscope but should be easier because linear operation is not required and the output voltage is modest.

What I would try is adding cascodes to the output current source and sink to add voltage gain and reduce the effects of Miller capacitance between the collector-base junctions.  That will at least increase the rise and fall time as much as possible.  Selective capacitive bypassing will reduce switching time and obviously you need to avoid saturation because of massively increased delay time.  Output cascode transistors will also allow using low voltage high speed transistors to drive the output cascode transistors.

If you are doing this with surface mount parts, then it may be time to consider fast saturated switches like the PNP MMBT5771 and NPN MMBT2369A.  RF parts are only suitable if baker clamped but should be considered also.  These are all low voltage yielding the requirement for output cascode transistors.

Clamping of the current output at a higher voltage may be needed instead of baker clamping because capacitance increases at low Vce.

Stick a small capacitor from the output to ground equal in value to your oscilloscope probe input capacitance to see how much the output capacitance is affecting the rise and fall time.

[PartialDischarge]

What you need looks more like the z-axis amplifier in a CRT oscilloscope but should be easier because linear operation is not required and the output voltage is modest.

Interesting reference. The old HP 182C uses common base amplifiers at the first stage.

[ali_asadzadeh]

Hero999 Thanks very much,I really appreciate your help in here:)

 David Hess thanks for your suggestions, would you share some circuits with us? of course I need to design the circuit with SMD parts.

[David Hess]

Interesting reference. The old HP 182C uses common base amplifiers at the first stage.

C10 is an example of the bypass capacitor which speeds up operation of the positive current source.

I think the Q5 common base amplifier is there to prevent the transimpedance output amplifier from going into saturation during blanking although it has other benefits.  It is a common feature in these designs.

Below the first example shows the Tektronix 7834 z-axis amplifier.  Q2206 and Q2216 are the common base input stages, Q2254 and Q2264 are the high voltage output cascode transistors, and Q2242 and Q2274 are the fast current source and sink transistors.  The current source and sink used reversed polarity transistors in this case with but they still work the same way with the difference in voltage between the emitters determining the currents.  I do not know why they reversed them and then used a PNP for error amplifier Q2236; other designs usually did not reverse them, used just two NPNs for Q2264 and Q2274 and no cascode on the PNP current source, and an NPN error amplifier.  I get the feeling after studying many Tektronix designs that they stopped optimizing when they had something which worked but in this case maybe this was the only way to get the needed performance with the parts they had.

This amplifier as shown would actually meet the requirements; it can generate at least a 24 volt output change in about 5 nanoseconds but it is also a linear amplifier and excessively complicated for the task of generating just pulses.  This is how you do it with early 1970s parts which is actually reasonable now because high performance high voltage transistors are not as common now as they used to be.

The second example below is from the Tektronix PG508 pulse generator.  Here C989 is the bypass capacitor connecting the top and bottom halves.  What these designs all have in common is the output transistors being in cascode configuration (common base) so the input signal is into the emitter instead of the base.  The paralleled common emitter output transistors are only there because this thing can drive 50 ohm loads to +/- 10 volts in less than 5 nanoseconds.  If driving a 50 ohm load to 24 volts is one of your requirements, then similar changes are needed.

Using a dedicated MOSFET driver is a good idea as they are often very fast but offhand I do not know of any which can operate with a 24 volt output.

[PartialDischarge]

Amazing designs, 40 years ago they knew things that most electrical engineers still don't know today , Bob was right...

[Zero999]

Interesting reference. The old HP 182C uses common base amplifiers at the first stage.

C10 is an example of the bypass capacitor which speeds up operation of the positive current source.

I think the Q5 common base amplifier is there to prevent the transimpedance output amplifier from going into saturation during blanking although it has other benefits.  It is a common feature in these designs.

Below the first example shows the Tektronix 7834 z-axis amplifier.  Q2206 and Q2216 are the common base input stages, Q2254 and Q2264 are the high voltage output cascode transistors, and Q2242 and Q2274 are the fast current source and sink transistors.  The current source and sink used reversed polarity transistors in this case with but they still work the same way with the difference in voltage between the emitters determining the currents.  I do not know why they reversed them and then used a PNP for error amplifier Q2236; other designs usually did not reverse them, used just two NPNs for Q2264 and Q2274 and no cascode on the PNP current source, and an NPN error amplifier.  I get the feeling after studying many Tektronix designs that they stopped optimizing when they had something which worked but in this case maybe this was the only way to get the needed performance with the parts they had.

This amplifier as shown would actually meet the requirements; it can generate at least a 24 volt output change in about 5 nanoseconds but it is also a linear amplifier and excessively complicated for the task of generating just pulses.  This is how you do it with early 1970s parts which is actually reasonable now because high performance high voltage transistors are not as common now as they used to be.

The second example below is from the Tektronix PG508 pulse generator.  Here C989 is the bypass capacitor connecting the top and bottom halves.  What these designs all have in common is the output transistors being in cascode configuration (common base) so the input signal is into the emitter instead of the base.  The paralleled common emitter output transistors are only there because this thing can drive 50 ohm loads to +/- 10 volts in less than 5 nanoseconds.  If driving a 50 ohm load to 24 volts is one of your requirements, then similar changes are needed.

What sort of transistors did they use? Did they have an especially high GBWP? One of the problems with fast rise/fall times is the transistors have very little current gain, I suppose it's possible to operate them above the GBWP, because they still have some voltage gain, even though the current gain is below one, but a very source low impedance is required.

Using a dedicated MOSFET driver is a good idea as they are often very fast but offhand I do not know of any which can operate with a 24 volt output.

Marco mentioned some MOSFET driver ICs which would fit the bill.

There are a couple gate driver ICs with sufficient voltage range and speed which could do this. IX 4426/27/28 for instance. You get rail to rail switching at relatively low static power consumption, this would be hard to replicate discretely.

[David Hess]

What sort of transistors did they use?

We can look it up since these designs were fully documented.  For the 7834 z-axis amplifier:

Q2242 Fast NPN   151-0411-00 6-5&13   2N5109
1.2GHz @15V@50mA 30V 3pF 400mA TO-39
Selected for rb'Cc<10ps

Q2274 Fast PNP   151-0417-00 6-19   Unknown
400MHz@10mA   4.5pF 40V 100mA TO-106

Q2274 Fast PNP   151-0220-00 6-3&15   2N3906
600MHz @20V@10mA 40V 4.5pF 200mA TO-92
Selected for rb'Cc<50ps

Q2254 Cascode PNP   151-0270-00 6-6&15 2N3495   151-0270-03 8-7 Selected
150MHz @10V@20mA 150V 3.5pF 300mA TO-39
Later selected for high hfe, high Vce, low Co, and high reliability

Q2264 Cascode NPN   151-0274-00 6-6&12 2N3501S   151-0274-01 8-8 Selected
150MHz @5V@50mA 150V 4.5pF 250mA TO-30
Later selected for high hfe, high Vce, low Co, and high reliability

Q2274 was original some bizarre specialized part but they replaced it with a 2N3906 selected for low collector-base time constant.  Tektronix also used these selected 2N3906s in their 100MHz 22xx oscilloscopes which seems odd to me because they had faster but still inexpensive parts available by then.

The cascode transistors were later selected for better performance and reliability.  I assume they had too many boards which were not fast enough or slowed down after being in service.  Would a transistor lose gain after days of burn in?

For the PG508 pulse generator:

Q990 PNP Cascode 151-0285-00 6-8&15 2N5160
Q1060 PNP Gain 151-0285-00 6-8&15 2N5160
500MHz @15V@50mA 40V 4pF 400mA TO-39

Q1070 NPN Cascode 151-0211-00 2N3866
Q980 NPN Gain 151-0211-00 2N3866
500MHz @25V@25mA 50V 3pF 400mA TO-39

Q1010 NPN Output Follower 151-0411-00 6-5&13 2N5109
1.2GHz @15V@50mA 30V 3pF 400mA TO-39
Selected for rb'Cc<20ps
Q1095 PNP Output Follower 151-0450-00 6-5&15 2N5583 Selected
1.2GHz @5V@150mA 35V 3.5pF 500mA TO-39
Selected for rb'Cc<10ps

These are all RF tranistors.  The PG508 uses a linear output stage because it allows variable edge rates which suggests looking at the 40 MHz output stage of the Tektronix FG504 function generator might be instructive.  HP had some similar instruments.  The big difference here compared to the 7843 z-axis amplifier is that the PG508 must be able to drive a 50 ohm load so they added a class-AB follower after the cascode output stage.

Did they have an especially high GBWP?

High voltage cascode transistors are not especially fast because that conflicts with high Ft.  The quest for higher bandwidth vertical CRT amplifiers in oscilloscopes was more about achieving higher deflection sensitivity in the CRT so that lower voltage higher bandwidth transistors could be used.  The 2N5109 is a real RF transistor.  The 2N3906 was selected for high Ft.

One of the problems with fast rise/fall times is the transistors have very little current gain, I suppose it's possible to operate them above the GBWP, because they still have some voltage gain, even though the current gain is below one, but a very source low impedance is required.

Bipolar transistors used in common base mode can have voltage gain yielding power gain after the current gain bandwidth product drops below one.

The input to the cascode transistors is very low impedance.  If necessary their output capacitance can be minimized by operating at high minimum Vce and they might be selected for low output capacitance anyway like transistors for the voltage gain stage of audio power amplifiers.  Those transistors are also getting hard to find which is why I mentioned the problem with availability of suitable parts being poorer than in the past.

Also note that a class-AB push-pull configuration delivers twice the transconductance because both sides are contributing to the output.

Without any output cascode stages or bypass capacitors, the 5 transistor 2N3904/2N3906 RS-232 level shifter I designed hit 50 nanoseconds without even trying although with a current output, this was sensitive to capacitive loading.  Actually looking at it now, I see that I used 2N4401s and 2N4403s which makes that even more impressive.  Unfortunately I don't know where I put the schematic.  It was just something I hacked together one afternoon and I choose an nontraditional but ordinary design for it just to see what kind of performance was actually possible with jelly bean parts.  When next I make one, I will include output cascode transistors.