Design advice: Clock/pulse generator output circuit

[ProtonFox]

Hi, I'm currently designing a simple clock generator using an STM32 and an Si5351. I actually don't really want to spend months on this project; this is why I chose parts I had laying around and could wire together on a breadboard. I managed to have a decent prototype which I am happy with. I only plan to use this project to have a somehow accurate clock generator to drive more elaborate pulse generators, test some RF mixers, trigger oscilloscopes, and testing digital circuits where my function generators are simply too slow. As a bonus having three independent outputs should be more than enough to cover my needs. Doesn't seem to be a big deal, right ?

That's all for the context! Now comes the part I need some advice. I want my outputs to be amplified and/or buffered so my device would have the standard 50 outputs and some level adjustment features. Nothing really fancy or accurate, maybe 1V, 3.3V and 5V peak-peak.

Here are the different solutions I had in mind:

1. use a broadband (DC to ~300 MHz or more) op-amp based circuit to perform this function. As some gain will be needed the op-amp GBW would be even higher. I first thought about this solution because I originally wanted to have a continuously adjustable level, but how to perform that? Not digital potentiometers (their bandwidth is not high enough), I would probably try to use a JFET as a voltage controlled resistor.

2. If I don't want to go fancy and use 2 or 3 fixed output levels, I would use some high speed logic gates to route the clock signals through different resistive attenuators. The result would be sent into a buffer/amplifier circuit (from solution 1 maybe). Looks probably easier to implement, but I'm concerned about gates output impedances and the extra propagation delays.

3. use the clock signal to drive a switching circuit, built from MOSFETs or an analog multiplexer. The switching circuit would "chop" a continuous voltage (which is easy to adjust) thus the output level would be adjustable. Problem: I haven't find yet an analog multiplexer fast enough with a small parasitic resistance. Maybe the discrete MOSFET solution would be enough (did some quick simulation tests with a 2N7002A with promising results ... but I need to have a P-channel complementary to have a true switch).

I'm also open for new ideas, that's also the reason I'm posting here today. So... in your opinion, what would be the best option or which downsides I haven't think of?

Finally, here are the specifications I try to achieve:
 - output frequency: 10 kHz (if needed I'm ready to start at 100kHz or even 1MHz) to 100 MHz (I can accept to go a bit lower, such as 80 or even 60 MHz if the signal gets to distorted).
 - rise/fall times: less than 4ns. According to the Si5351 datasheet (okay... some people say it's not always reliable) I should have rise/fall times less than 1.5ns, but it looks like it will be difficult to gain some extra nanoseconds.
 - output impedance: 50

Thanks in advance!

[shapirus]

Problem: I haven't find yet an analog multiplexer fast enough with a small parasitic resistance. Maybe the discrete MOSFET solution would be enough

Well, a possible candidate may be currently discussed right there in the neighboring topic: SN74LVC1G3157. Fast (sub-ns), but only SPDT. You could cascade a few, though, if you need to switch between more than two outputs.

[ProtonFox]

Oh, that part looks like a good candidate for this function! It also seem to be dirt cheap and readily available.

To be honest I don't care about switching between several outputs, because I plan every clock output to have its independant analog output stage, so having a SPDT switch is not a problem.

[MasterT]

I 'd think about option 3 "in reverse", when not to use analog switches (high calacitance) but "clip" output waveform to adjustable high / low level using RF or schottky diodes.

[ProtonFox]

That's a smart idea. Obviously I'd drive this clipping circuit with an amplified signal so I can have my 5V output level, but this seems feasible. It is also easier to simulate (no "black box" or encrypted models from TI I can't use  ) so I can try this out quickly.

[ProtonFox]

Okay, so after playing around with NgSpice testing these two solutions, here are my conclusions:

Using an analog switch such as the suggested SN74LVC1G3157 could work if capacitance between inputs and outputs is kept in the order of tens of pF. This needs to be taken carefully though, because I got these results from a really crude model (a switch with internal on resistance in parallel with a parasitic capacitance) because I could not get an actual model of the IC.

Trying to clip the output works as long as the diode threshold voltage is low enough and parasitic capacitance is negligible, which is actually pretty easy with RF schottky diodes. I managed to get an output range of 0.8V to 5V, open-circuited output, assuming a 5V output square wave from an amplifier. That level tuning is not linear but it was expected with diodes. Output seems to be more distorted especially at lower levels when adding a properly terminated transmission line.

I didn't do some serious engineering, I did that mostly to figure out which idea works the best. I'm rather happy with the two options, so now what do I plan ?

From a design perspective, the diode clipper is a bit more complicated to design and expensive to build: the clock signal needs to be amplified, the tuning voltage needs to be generated whether by fixed preset values or a DAC, and of course the diode clipping circuit. The analog switch circuit would need directly driven by the clock signal and chop the output of a DAC (with some proper filtering/decoupling circuits of course).

To be fair, I want to prototype the analog switch circuit because I want to see how well they perform at high frequencies. If I can get interesting results from them, it would be something interesting for RF projects, for example. If I'm not happy with the result, I'd spend more time designing a proper diode clipping circuit afterwards.

Sounds like it's gonna be a longer journey than I thought, after all. 

[shapirus]

How do you implement variable level with diode clipping?

[ch_scr]

You could get the continuously variable output level with a logic gate (buffer or inverter) supplied from a variable voltage.
It's output impedance might change with supply level, but tiny gates (single inverter like 74lvc1g04) can tolerate higher drive than the supply no problem.
This in combination with (analog switch selected) attenuators could give a wide range of continuously variable output levels. Maybe no need to chain the analog switches, then.
Don't think the gate delay is of much relevance here? To drive transmission lines, you want a good match to 50R to make the square edges look clean.
This can be done by always having some minimal attenuation to improve match.

[David Hess]

3. use the clock signal to drive a switching circuit, built from MOSFETs or an analog multiplexer. The switching circuit would "chop" a continuous voltage (which is easy to adjust) thus the output level would be adjustable. Problem: I haven't find yet an analog multiplexer fast enough with a small parasitic resistance. Maybe the discrete MOSFET solution would be enough (did some quick simulation tests with a 2N7002A with promising results ... but I need to have a P-channel complementary to have a true switch).

Instead of MOSFETs and voltage switching, use bipolar transistors and current switching, or fast diodes and current switching.  The output level is then set by the easy to control DC current source driving the emitters of the differential pair, or driving the diodes.

[ProtonFox]

Thanks for the other suggestions ch_scr and David Hess,

How do you implement variable level with diode clipping?

It's actually pretty easy to have a basic variable diode clipping by changing the reverse bias of a diode. I attached the schematic I made to test this idea, which was probably what MasterT had in mind.

You could get the continuously variable output level with a logic gate (buffer or inverter) supplied from a variable voltage.

Well, that's basically what the analog switch circuit does, except it is one of the inputs of the switch which is supplied by a variable voltage.

To drive transmission lines, you want a good match to 50R to make the square edges look clean.
This can be done by always having some minimal attenuation to improve match.

That's true, one of my biggest priority is to have a nice 50 match to avoid reflection problems and stuff. I'm concerned about having a nice, controlled output impedance. Output impedance of logic gates is almost never mentionned in datasheet, as they seem to be targeted to digital engineers; parasitic resistance of analog switches is given but I don't know if it is actually consistent between batches or even chips themselves. But having some attenuation at the output might be a way to be more confident about output matching.

Instead of MOSFETs and voltage switching, use bipolar transistors and current switching, or fast diodes and current switching.  The output level is then set by the easy to control DC current source driving the emitters of the differential pair, or driving the diodes.

That's an interesting approach, it also looks like it was used in some pulse generators such as the Philips PM5715, which I borrowed the idea of chained attenuators in the output circuit. I may try it too.

[fourfathom]

Here's the output stage I did for a triple-output Si5351 design.  I wanted 0-3.3V open circuit output, and 50 Ohm output resistance (0-1.6V output when terminated).  This design also has selectable AC-coupling on the output.  This schematic shows two buffers, this lets me use two different footprints depending on which package is available.  I do *not* install both parts.
The frequency response is reasonably good up to 100 MHz.  In the KHz range there is a bit of droop caused by the ferrite choke in the supply line, but the 10uF bypass cap works well enough here.

[shapirus]

What's that AQY2B25X part? Is there a typo? Google returns literally nothing when doing exact match search.

(it's clear that it's some optical switch, I wanted to find its datasheet)

[fourfathom]

What's that AQY2B25X part? Is there a typo? Google returns literally nothing when doing exact match search.

(it's clear that it's some optical switch, I wanted to find its datasheet)

That an AQY282SX from Panasonic (the schematic text is hard to read). They call it a "PhotoMOS Relay".  There are a few variants, some with higher on-resistance and lower capacitance (I use these in High-Z  circuits) The one here has a lower on-resistance and of higher in/out capacitance (which doesn't matter here).  The isolation voltage is 1500V and the switch max voltage is 60V or better.  Useful parts and LCSC usually has them in stock!

[PCB.Wiz]

3. use the clock signal to drive a switching circuit, built from MOSFETs or an analog multiplexer. The switching circuit would "chop" a continuous voltage (which is easy to adjust) thus the output level would be adjustable. Problem: I haven't find yet an analog multiplexer fast enough with a small parasitic resistance.

That's looking the best approach.
It is a trade off, and there are many Analog Switches that head into sub-ohms territory, but they have large capacitance.

When looking into this trade-off not long ago, I found the USB switches were good at reasonably low resistance without killer parasitics.

Most are 3v3, so I'd suggest a LVC buffer pair as above for a fastest / dedicated 5V output, and then USB switches like Toshiba TC7USB40FT or TC7USB42FT in TSSOP14 etc for your adjustable output

TC7USB40FT  claim
Supply voltage:  VCC = 2.3 to 4.3 V
Switch terminal ON-capacitance: CI/O = 5 pF Switch ON (typ.) @VCC = 3.3 V
ON-resistance: RON = 4.5 Ω (typ.) @VCC = 3 V, VIS = 0 V
RON flatness: RON(flat) = 1.3 Ω (typ.)@VCC = 3 V
BW  RT = 50 Ω, CL = 0 pF,  1500MHz

Or parts like NX3DV221 come in XQFN10 HVSON10, which claim 5.5V analog levels,
6 Ohms (maximum) ON resistance
6 pF (typical) ON-state capacitance
High bandwidth (1.0 GHz typical)

That x221 series looks to be multi sourced, and claims a charge pump so they can use a single N-FET for lower parasitic C.
I did not see typical changeover times, so you may need to experiment.

Addit: I see the OnSemi FSUSB42MUX looks a good fit. Specs typical ton/toff of 12,13ns (3V) so has low skew, and is 3.9 Ohms, 3.7pF, 720MHz
RECOMMENDED OPERATING CONDITIONS  VCC Supply Voltage 2.4 ~ 4.4 V  VSW Switch I/O Voltage −0.5 ~ 4.5 V


Another candidate would be 74CB3Q3257PW - has higher Con, but does spec a faster sel tpd  so you could try that to see what bites first : Con or Tpd ?
It looks to run the charge pump all the time, other parts gate the charge pump on OE. It also specs 5V Analog path switch from 3v3 Vcc

Finally, here are the specifications I try to achieve:
 - output frequency: 10 kHz (if needed I'm ready to start at 100kHz or even 1MHz) to 100 MHz (I can accept to go a bit lower, such as 80 or even 60 MHz if the signal gets to distorted).

You should be able to have the full Si5351A register range, ~3kHz to ~225 MHz

[David Hess]

Instead of MOSFETs and voltage switching, use bipolar transistors and current switching, or fast diodes and current switching.  The output level is then set by the easy to control DC current source driving the emitters of the differential pair, or driving the diodes.

That's an interesting approach, it also looks like it was used in some pulse generators such as the Philips PM5715, which I borrowed the idea of chained attenuators in the output circuit. I may try it too.

Tektronix used it on most of their pulse generators.  They used the diode version for vertical channel switching in oscilloscopes which gives some idea of the potential performance.

[ProtonFox]

Didn't expect so many replies, thanks everyone. It's good to know many approaches as I struggled finding interesting solutions when doing my research.

I don't think I can achieve an adjustable output level using your method fourfathom, but maybe I'll borrow your schematic if I can't reach this goal. I sometimes wonder if it's actually worth the hassle to have an adjustable output anyway. Well, TTL high level is 5V so 3.3V may be a bit too low. On the other hand I don't want to fry low voltage CMOS circuits (even if I'm more of an analog/RF guy so I don't encounter that kind of stuff very often). So I'll keep trying to have a somewhat working adjustable level, even if the level range is narrow nor accurate.

Right now I'm still on the analog switch method (schematic attached) using SN74LVC1G3157 but I may look at the references suggested by PCB.Wiz. By the way, will the switch mess up the op-amps output or something like that?

Tektronix used it on most of their pulse generators.  They used the diode version for vertical channel switching in oscilloscopes which gives some idea of the potential performance.

Have you got any schematic? I don't know Tektronix products very well apart from their oscilloscopes so I don't know where to search.

Edit: nevermind, I found everything I needed on TekWiki.

[PCB.Wiz]

Right now I'm still on the analog switch method (schematic attached) using SN74LVC1G3157 but I may look at the references suggested by PCB.Wiz.
By the way, will the switch mess up the op-amps output or something like that?

Yes, the opamp output impedance increases with frequency, whilst you need a very low impedance source up to 100's MHz.
The LM324 is a also a bit light for current source, if you want full swing at low frequencies, the peak current needs to come from the driver.
The LM324 will also not do rail-rail output, but you may be ok with less ?

[PCB.Wiz]

Right now I'm still on the analog switch method (schematic attached) using SN74LVC1G3157 but I may look at the references suggested by PCB.Wiz. By the way, will the switch mess up the op-amps output or something like that?

It's an interesting problem, so I did a quick check.

You are basically building a programmable power supply, that needs to tolerate large transient load swings, say 5-50mA

The LM324 fails in at least 3 areas - current is too low, and it is not Rail to Rail IO

a) You can get opamps with enough drive, but they need decoupling capacitance on the regulated rail, so you need high-C-load  R/R/C type output design.

b) A discrete NPN diff + PNP output works quite well, the gain is low enough it can be stable at uF loads. The single large C-Load gives a single pole.
However, the simplest design only works down to ~ 700mV.
A change to PNP Diff + Current Mirrors + PNP output managed close to GND~4.2V input, thus a gain of 1.25  can give 5V output swing from 4V input.
After the range is expanded, you are up to 3 matched pairs SOT26 and a PNP, so the BOM creep here probably make this less attractive.

c) Since you have a DAC, small adjustable LDOs could be used instead.
They are designed and spec'd for ADJ power, and are single package and cheap.
The better ones are spec'd for fast transient response, tho you have one-per output so are more tolerant of load step response.

 

[ProtonFox]

Ah yes, did not think about the transient currents which can be quite high in this case. I first used the LM358/LM324 because I assumed it would be able to deliver enough DC current and I actually don't care about the accuracy of my levels, so I thought I could just slap a jellybean op-amp. Also turns out that it's on the "sensitive path" of fast transients. 

It doesn't look like it's worth using better op-amps to perform this function, so I'm gonna dig further the idea of the simple discrete BJT output circuit. I don't want my output level range ot be wide nor accurate; I'll use a more sophisticated pulse generator (I do have one already) driven by this circuit if I have special needs anyway. If I can achieve some control between ~1V up to 5V, mostly to cover most basic needs in the lab. I could maybe even ditch the DAC and use a more minimalistic approach, but this will be determined later.

I don't know much about LDOs and I'm kind of concerned about their output impedance/capacitance at high frequencies.

[ProtonFox]

Thinking more about how I'm going to use this project and how much time I actually want to spend working on it, I begin to believe it's a bit overkill to make an elaborate output circuit with an adjustable output level. Especially if I don't want it to be really accurate or having a wide output level range. I'll mostly use 2, maybe 3 outputs levels anyway. So this brings back the option 2 from my first post.

I'd prefer to have two buffers, let's say one that can output 3.3V and one that can output 5V. Their outputs would go into an analog switch and route the buffered/amplified signal to a BNC output (well, through a series resistance for output impedance matching). I'll then focus on designing a proper output stage without having to bear the hassle of having more design considerations. Even routing the signals properly, selecting the best components to perform this function, won't be trivial.

I'll see if I need a more complicated output stage in the future; all your design ideas will still be useful, especially to other people building pulse generators. That's the beauty of forums. 

[PCB.Wiz]

It doesn't look like it's worth using better op-amps to perform this function, so I'm gonna dig further the idea of the simple discrete BJT output circuit. I don't want my output level range ot be wide nor accurate; I'll use a more sophisticated pulse generator (I do have one already) driven by this circuit if I have special needs anyway. If I can achieve some control between ~1V up to 5V, mostly to cover most basic needs in the lab. I could maybe even ditch the DAC and use a more minimalistic approach, but this will be determined later.

I looked into a simple level shift + modest gain, applied to a NPN diff + PNP BJT design and it is likely 'good enough'.
It is well behaved with impulse current loads, and output impedance about 0.4 ohms.
Gives 123.43mV to 4.912V OUT range DC average. You can see the ripple increases at the extremes as it hits saturation.

 

I don't know much about LDOs and I'm kind of concerned about their output impedance/capacitance at high frequencies.

A benefit of a well behaved BJT design, is you get clean edges on the square wave, so you can use it to check impulse ringing on equipment.
LDOs tend to have some impulse spikes, as their lower output impedance eventually corrects.

With 3 outputs from Si5351A, you could have buffers as one fixed at 5V, one fixed at 3v3, and one adjustable.

[fourfathom]

Since you want a square-wave output, is there any reason not to use the NL27WZ17, as I showed a few posts above, or the similar-performance SN74LVC2G04 Dual Inverter Gate?  Feed the Vdd pin from a variable or selectable voltage (1.8V to 5V) and you get the square-wave amplitude you want.  Add a series resistor to establish the 50 Ohm output impedance (I paralleled two buffers and output resistors to reduce the effect that any gate impedance variability had on output Z.  I assume that the gate output Z will vary somewhat with Vdd, but unless you need high precision a compromise resistor value should work.

If you are driving the gate directly from the 3.3V Si5351 output you will want to add a buffer / translator to jack that signal level up to +5V before driving the NL27WZ17 (if that will be operating at +5V).  The NL27WZ17 can accept +5V logic-level inputs even when running at Vdd = 1.6V.

[PCB.Wiz]

Since you want a square-wave output, is there any reason not to use the NL27WZ17, as I showed a few posts above, or the similar-performance SN74LVC2G04 Dual Inverter Gate?  Feed the Vdd pin from a variable or selectable voltage (1.8V to 5V) and you get the square-wave amplitude you want.  Add a series resistor to establish the 50 Ohm output impedance (I paralleled two buffers and output resistors to reduce the effect that any gate impedance variability had on output Z.  I assume that the gate output Z will vary somewhat with Vdd, but unless you need high precision a compromise resistor value should work.

Yes, that can work for higher fixed supplies. As you mention, the lower Vcc drops the drive.
An Analog Switch generated square wave does not suffer from drive droop at lower voltage, meaning it can work < 2V  and even <1V

If you are driving the gate directly from the 3.3V Si5351 output you will want to add a buffer / translator to jack that signal level up to +5V before driving the NL27WZ17 (if that will be operating at +5V).  The NL27WZ17 can accept +5V logic-level inputs even when running at Vdd = 1.6V.

A purist might use a 'T' translating device for split Vcc work, but some vendors (Diodes,Nexperia) do offer a 74LVC2G34, that specs VIH of 2V at 3.6V Vcc, (55.5%) rather than the lazier 70% from decades ago.
That shifted threshold translates to 2.85V max, on a 5V operating part.

[ProtonFox]

Sorry for the late reply, I didn't had time to reply or work on this project.

Since you want a square-wave output, is there any reason not to use the NL27WZ17, as I showed a few posts above, or the similar-performance SN74LVC2G04 Dual Inverter Gate? Feed the Vdd pin from a variable or selectable voltage (1.8V to 5V) and you get the square-wave amplitude you want.

I'm not a fan of this method because I don't find it really proper. Output impedance and slew rates tend to vary quite drastically when changing the supply voltage of these chips as you said.

After some additional thinking, I decided to give a last chance to the analog switch method by setting up a more serious simulation. Moreover PCB.Wiz gave some useful hints I want to try, and I'm not against a little transistor circuit design exercise.

I'll show results in the next days if I can achieve something nice from it. Otherwise I'll just go back to the simplified design.

[shapirus]

I'm not a fan of this method because I don't find it really proper. Output impedance and slew rates tend to vary quite drastically when changing the supply voltage of these chips as you said.

Slew rates, maybe. Output impedance... I wouldn't be so sure before testing it in actual hardware.