High-Z wideband buffer amp for 50 ohm CRO inputs

[GK]

I'd like to use 10:1 10M probes with my Tek 7904 500MHz oscilloscope whose wide-bandwidth vertical plugins sport 50-ohm inputs. I've elected to build a 4-channel buffer amplifier for this purpose. I'm aiming for a bandwidth of at least 250 MHz. While this is only half the bandwidth of the 7904 mainframe I feel that there is no point going any more exotic as the achievable bandwidth with 10:1 10M probes limited. After a couple of hours of head scratching this evening here is my preliminary schematic.

The unit to be constructed will have four identical channels. A channel is comprised of a 1M input termination, AC/DC coupling capacitor/switch(relay) and input protection followed by an RF JFET buffer, followed by a x2 Av op-amp based amplifier with a 50-ohm output Z for driving the scope input via a length of connecting coax. 

The ADA4899-1 output will clip/limit at about 3V peak driving the combined 100 ohm load + 300 ohm feedback resistance. This gives a full scale input of 15V peak with a 10:1 probe, though in use an upper limit of 10V peak would be considered the useful maximum. I feel that this is more than adequate as high frequency solid-state circuits seldom have ac amplitudes greater than 20Vpp.

One thing that has to be accounted for is the positive DC offset present due to the gate-source voltage of the JFET. I've elected to use a small 8-bit PIC uC (only 7 IO pins required) in combination with a four-channel 12-bit serial DAC to auto-null each channel immediately after power-on.

At first I considered using a uC with an internal ADC and MUX to measure the DC-level output of each channel during the auto-null, but that would require level-shifting to accommodate the ADCs input range, and that would have a significant degree of inaccuracy necessitating manual trimming for each channel. So as an alternative I decided to use an external MUX controlled by the uC feeding a zero-crossing comparator rather than an ADC.

The uC will be programmed with a simple successive approximation routine - each DC-null will take twelve bit-toggles with the final status for each bit determined by the output of the comparator. The end result of this is that the DC offset will be auto corrected to a couple of mV (which is more than adequate) without the need for any manual tweaking or precision components. During nulling the uC activates relay K2 to short the JFET input to ground.

I can't see anything immediately bothersome or not straight forward, though I have yet to select an suitable RF JFET. Require something with a min. Idss on the order of 10mA; currently considering the MMBFJ309 as I notice this is what Rigol use in the DS2027A, though I'd prefer something with a smaller worse-case Vgs.

Comments?




 

[tggzzz]

I'd like to use 10:1 10M probes ... I'm aiming for a bandwidth of at least 250 MHz.

Before you go too far, don't forget that there is no such thing as a 10:1 10Mohm probe.

Typically, at 250MHz their input impedance will be dominated by the tip capacitance; 15pF@250MHz => 42ohms. Use a FET or "low impedance Z0" probe. Or just a terminated 50ohm cable

[GK]

Before you go too far, don't forget that there is no such thing as a 10:1 10Mohm probe.

Typically, at 250MHz their input impedance will be dominated by the tip capacitance; 15pF@250MHz => 42ohms. Use a FET or "low impedance Z0" probe. Or just a terminated 50ohm cable

I am aware of this. I have a 50-ohm terminated cable whenever I use the vertical channel inputs of 7904 directly. This unit is to serve a different purpose. Perhaps I should have written that I intend to give my 7904 standard 1M-ohm inputs, for convenience.

Additionally I have removed my "edit" note and I now have the correct schematics up. Who needs to go to bed on a Friday night anyhow?
 

[GK]

Ugh... Forget about the ADA4899-1. Nice bandwidth but its slew rate is lousy. Will have to look for something much better.

[GK]

.... more issues......

Assuming that the source of the JFET sees a total load capacitance of 10pF. At 100MHz 1Vpeak this will require a peak drive current of a tad over 6mA. That is near the bias current of the JFET and will result in massive distortion.

A practical limit would be 100mV peak signal amplitude at the JFET. Anyone know what signal amplitude the Rigol front-end runs on its JFET buffers?

This would then require a frequency-compensated 10:1 attenuator on the input side of the JFET to maintain the desired input signal handling capacity.
However 10 times less signal amplitude at the output means that I no longer have slew-rate limitation issues with the ADA4899-1. At 200mV peak output (double to account for the 6dB loss of driving the terminated 50 ohm scope inputs) the specified 310V/uS slew rate will accommodate a power bandwidth of 310e6/(2pi*200mV) =  247MHz..... although that is just adequate for the bandwidth target.... I would prefer that the full power bandwidth computed from the slew rate limitation be twice (for a decent safety margin) the achieved small-signal bandwidth.   

However the additional 20dB attenuation turns a 10:1 probe into a 100:1 probe, limiting me to only 1V per division maximum sensitivity (scope vertical plug-in does 10mV/div). Id need to make the input attenuator switchable to get this back to a more useful 100mV/div.

 

[lukier]

Interesting project, maybe it could be adapted to work like TCA-1MEG for those using TDS7404 and other TekConnect input scopes (those TCA-1MEG are pretty expensive). Also spectrum analyzers that have 50 Ohm inputs. 

[Someone]

With so little gain in that front end you can get the bandwidth and swing from opamps alone. Here is is a much higher gain design doing the same impedance conversion:
https://www.eevblog.com/forum/projects/diy-100mhz-differential-probe/msg938343/#msg938343

[T3sl4co1l]

BF862 is quite popular.  2N5486 might be okay too, but the LF noise is uncontrolled and may be visibly poor.

Speaking of, there's a diplexer arrangement that's popular for lower noise and wide offset; I think Rigol uses it too?  (AC couple input to JFET gate; 100M from JFET gate to op-amp; op-amp servos JFET to keep DC(output) = DC(input).)  Have you considered this, or just figure it's not worth it?

(You'll probably want attenuators too, but I suppose that's just not shown.)

Tim

[Alex Eisenhut]

http://w140.com/tekwiki/wiki/282

[GK]

With so little gain in that front end you can get the bandwidth and swing from opamps alone. Here is is a much higher gain design doing the same impedance conversion:
https://www.eevblog.com/forum/projects/diy-100mhz-differential-probe/msg938343/#msg938343

I'm going to need something with at least double the slew rate of the ADA4899-1, so probably a CFA. These high speed (>600MHz GBWP) op-amps have high input bias currents that don't make for convenient termination into 1M. The datasheet for the THS306(1/2) you used specifies a typical input bias current for the non-inverting input of +/-6 A !. Though this is obviously a typo and the unit should be uA. That's +/-6V DC offset when terminated into 1M. I also don't think that many op-amps with =>600MHz GBWP would be very stable with a source termination of 1M||9M. I think the JFET buffer has to stay.
 

[GK]

OK, reviewing the design spec..............

The unit will be built just as shown in the preliminary schematic, with the exception of a better op-amp substituted for the ADA4899-1 and a relay-switched 20dB attenuator added to the input circuit. This is as complicated as I'm willing to make the design, which is simply meant to be a moderately convenient probe adapter and not a full blown analogue CRO front-end.

[Someone]

With so little gain in that front end you can get the bandwidth and swing from opamps alone. Here is is a much higher gain design doing the same impedance conversion:
https://www.eevblog.com/forum/projects/diy-100mhz-differential-probe/msg938343/#msg938343

I'm going to need something with at least double the slew rate of the ADA4899-1, so probably a CFA. These high speed (>600MHz GBWP) op-amps have high input bias currents that don't make for convenient termination into 1M. The datasheet for the THS306(1/2) you used specifies a typical input bias current for the non-inverting input of +/-6 A !. Though this is obviously a typo and the unit should be uA. That's +/-6V DC offset when terminated into 1M. I also don't think that many op-amps with =>600MHz GBWP would be very stable with a source termination of 1M||9M. I think the JFET buffer has to stay.

The DC path is through the precision amplifier so you do not see the bias current of the CFA input pins, and for your application the entire loop could have a DC servo to eliminate the micro controller and DAC/ADC you propose. Just another way of getting the impedance transform, let us know how you go with your plans.

[GK]

A DC servo is out of the question as that would give the unit a high-pass frequency response; ie a frequency response that does not extend to DC (Note that in the Rigol front-end the LF amplifier is not a DC servo)

Yes another way of building the front end is with a JFET low-bandwidth op-amp bypassed for HF, but that has other drawbacks and isn't any simpler (it *still* requires a high input impedance buffer/amplifier for the combined LF/HF path).


In my  proposed circuit the only active component loading the 1M input network is the JFET. This "minimalism" will help in keeping the parallel (shunt) input capacitance to a minimum, which is preferable as I'd like to unit to be compatible with low-c, high-bandwidth probes.
 

[bson]

Ugh... Forget about the ADA4899-1. Nice bandwidth but its slew rate is lousy. Will have to look for something much better.

How about the BUF602?

[GK]

Ugh... Forget about the ADA4899-1. Nice bandwidth but its slew rate is lousy. Will have to look for something much better.

How about the BUF602?

Unfortunately I need an op-amp configuration as I require a gain of 2 and means of shifting the DC offset. AD8001 looks like it will do the job.
 

[GK]

This is what the design currently looks like. Relay K1 is used the switch the 20dB input attenuator in and out. I'm aiming for an input shunt capacitance of 18pF. This will be layout dependent to some degree, so it will be padded up if required with the capacitor labeled "S.O.T". Trimmer CV1 is used to equalize the input capacitance when the attenuator is switched in, so that the probe calibration holds. CV2 is used to adjust the flatness of the frequency compensation.

The 4-channel 12-bit DAC for the micro-controlled offset nulling will be Burr Brown (now TI) part # DAC7614.



EDIT: Oops, I omitted the AC/DC coupling capacitor/relay-switch immediately after the input BNC.

[T3sl4co1l]

Hm, should K2 not be SPDT so it doesn't short the input?

What's the bias below the JFET doing?  It just appears to be a double cascode on top of 680+68 ohms, so the dual and the diode aren't doing anything..

Also something mumbled about superfluous bypass caps, but whatever.

Tim

[joeqsmith]

Have you looked at some of the newer op-amps?    I wanted to do something similar.  DC-???  1M input 50 output.  +/-30V input w/ protection.  Video showing a TI eval board.   I played with a couple of different ones.

[Cerebus]

May I recommend a read of the article "Signal Conditioning in Oscilloscopes and the Spirit of Invention" by Steve Roach of Tektronix. Basically it's a high level tutorial on designing 1M scope inputs from someone who does it for a living - highly recommended. It'll give you some ideas about FETs but it might cause you to rethink some of how you're planning to do this.

[GK]

Hm, should K2 not be SPDT so it doesn't short the input?

What's the bias below the JFET doing?  It just appears to be a double cascode on top of 680+68 ohms, so the dual and the diode aren't doing anything..

Also something mumbled about superfluous bypass caps, but whatever.

Tim

What? As previously explained K2 is for shorting input under uC control while auto-zeroing of the output takes place.

Routing the input through another set of relay contacts will only add to the input capacitance and there is no need to do so in any case. When K2 is activated there is no danger from input currents flowing in via the 9M series resistance of a 10:1 probe or the few puff of compensation capacitance in parallel with it.

The current source is designed to be highly stable to prevent DC drift. The diode-connected transistor of the dual temperature compensates the other acting as the current source - thus the voltage across the 680 + 68 ohm remains highly stable. The -10V rail will voltage-track the -5V rail with temperature as the (SMD LM337L) regulator for the -10V rail will use the -5V rail as a "ground" reference. 

The current source is cascoded because the dual BJT has high collector C and capacitive loading on the JFET source has to be kept as low as practical. So the current source is cascoded with a RF BJT. Not very complicated or difficult to understand, IMO.

[GK]

May I recommend a read of the article "Signal Conditioning in Oscilloscopes and the Spirit of Invention" by Steve Roach of Tektronix. Basically it's a high level tutorial on designing 1M scope inputs from someone who does it for a living - highly recommended. It'll give you some ideas about FETs but it might cause you to rethink some of how you're planning to do this.

I have the article and there is nothing wrong with the way I have implemented the 1M input network.

Thanks.

[GK]

BF862 is quite popular.  2N5486 might be okay too, but the LF noise is uncontrolled and may be visibly poor.

The BF862, despite being marketed for AM radio front-ends, is popular in audio (phono preamps, etc) because it has very low voltage noise (<1nV typical) and a very low (for an RF JFET) 1/f corner (<1kHz). It also has high junction capacitances compared to typical VHF/UHF JFETs, ruling it out as a contender.

The higher "LF noise" of typical VHF/UHF JFETs isn't an issue anyway when the input source is 1M||18pF.

[Cerebus]

May I recommend a read of the article "Signal Conditioning in Oscilloscopes and the Spirit of Invention" by Steve Roach of Tektronix. Basically it's a high level tutorial on designing 1M scope inputs from someone who does it for a living - highly recommended. It'll give you some ideas about FETs but it might cause you to rethink some of how you're planning to do this.

I have the article and there is nothing wrong with the way I have implemented the 1M input network.

I didn't say there was, just that the article might make you rethink - because there's such a lot of very useful info in it. Why so snippy?

[GK]

Have you looked at some of the newer op-amps?    I wanted to do something similar.  DC-???  1M input 50 output.  +/-30V input w/ protection.  Video showing a TI eval board.   I played with a couple of different ones.

I can't watch the vid (internet connection limitations), but I think the op-amp approach is practical if you can live with a high ratio of fixed attenuation at the input. That way the impedance "seen" by op-amp can be kept low enough to prevent instability. A 1M input termination for an op-amp with a GBWP approaching 1GHz though just doesn't seem to be conductive to stable operation. In applications such as video line drivers (gain +2), where the datasheets actually go into layout and stability requirements for these kind of op-amps, it's typically recommended to terminate the input with the low impedance (75 ohms) as close as practical to the op-amp input.

It would be nice if it were so simple, but I have yet to see a commercial DSO front-end design of several hundred MHz bandwidth in which the 1M input is successfully buffered with a low-input-Ib op-amp voltage follower.

Though if anyone can present an example I'm happy to be proven wrong.

[T3sl4co1l]

What? As previously explained K2 is for shorting input under uC control while auto-zeroing of the output takes place.

Oh right.. I should pay attention more

The BF862, despite being marketed for AM radio front-ends, is popular in audio (phono preamps, etc) because it has very low voltage noise (<1nV typical) and a very low (for an RF JFET) 1/f corner (<1kHz). It also has high junction capacitances compared to typical VHF/UHF JFETs, ruling it out as a contender.

Not necessarily.  Phil Hobbs is quite fond of them in photodiode TIAs and esoterica.  The capacitances go down because it's a follower; they go down further still if you add bootstrapping, which isn't too big of a deal (a BJT follower into the drain, with a resistor divider for bias, and a cap coupled to the source, is the simplest starting point).

After killing off most of the capacitance, the high transconductance is a big payoff.

Tim

[GK]

I'm not going to bootstrap Vds to save maybe one pF of input capacitance when I can just select one of many parts that will give me one pF less input capacitance to start with.

[joeqsmith]

Attempted to send you a PM but "User 'GK' has blocked your personal message."

I'm curious what limitations you have.  Bandwidth, or possibly blocked due to the music?    Info may help me with future videos.

Thanks

"I can't watch the vid (internet connection limitations), but I think the op-amp approach is practical if you can live with a high ratio of fixed attenuation at the input."

[joeqsmith]

Maybe this is of interest:

http://www.amazon.com/High-Z-Spectrum-Analyzer-Amplifier-FETAMP1/dp/B00MEVDX70

Some of the new amps are not too bad. 

[GK]

Hmm... I've put JFET-input op-amp ADA4817-1 on my next order.

[joeqsmith]

Video was an OPA6xx on one of their eval boards.   Be interesting to see what you come up with.

Hmm... I've put JFET-input op-amp ADA4817-1 on my next order.

[GK]

Be interesting to see what you come up with.

The OPA657 looks nice, but doesn't have the slew-rate and +2 gain bandwidth required. The ADA4817-1 however looks to be the bee's knees as far as FET-input op-amps go in terms of bandwidth and slew rate. On +/-5V supplies the ADA has a specified slew rate of 870 V/us, but that is with a 4V input step which will massively overload the input stage and deliver the full tail current to the internal compensation capacitance for the longest duration of time. For a gain of +2 set with a pair of 200 ohm feedback resistors, it appears from the small and large signal response plots provided in the datasheet that the effective slew rate for sinusoidal signals is in the order of 600-650 V/us and the -3dB bandwidth 400 MHz with no HF peaking. That will support an output sine of 0.4V peak at 250 MHz.

That is more than adequate performance so it looks like there is no point in pursuing the discrete implementation. Sad, as I was quite enamored with my JFET buffer and auto-zeroing scheme, lol.

The evolved circuit, now in its third iteration, appears below. I've shifted the AC/DC coupling relay and cap to after the input attenuator; I decided that I didn't want the additional path inductance between the probe and the shunt termination capacitance. From experience I'm still a bit paranoid about terminating the input of such a high speed op-amp to such a high source impedance. The 1.8pF cap and the 100R resistor in the schematic will be put right at the non-inverting input pin. 1.8pF at 400MHz = 220 ohms, which should nicely keep the input terminated into a low Z at frequencies that could otherwise prove troublesome for pick-up and regeneration. The 100R will neuter any series resonant peaking between the input C and the net trace/component inductance of the loop closed by the shunt termination capacitance over at the input attenuator. Well that's the theory anyway. Proof will come in solder. I'm starting the board layout for a prototype channel now. 
 




   

[joeqsmith]

I was using a 53 that uses the built-in network.   Was looking to get 100MHz with a fixed gain and 60Vp-p.  Video compared results of coax vs a 10X w/driver from DC to 400 or so MHz.    4V step for the 53 is 2675V/us with a gain of 2V/V and 100 ohm load.   

The OPA657 looks nice, but doesn't have the slew-rate and +2 gain bandwidth required. The ADA4817-1 however looks to be the bee's knees as far as FET-input op-amps go in terms of bandwidth and slew rate. On +/-5V supplies the ADA has a specified slew rate of 870 V/us, but that is with a 4V input step which will massively overload the input stage and deliver the full tail current to the internal compensation capacitance for the longest duration of time. For a gain of +2 set with a pair of 200 ohm feedback resistors, it appears from the small and large signal response plots provided in the datasheet that the effective slew rate for sinusoidal signals is in the order of 600-650 V/us and the -3dB bandwidth 400 MHz with no HF peaking. That will support an output sine of 0.4V peak at 250 MHz.

That is more than adequate performance so it looks like there is no point in pursuing the discrete implementation. Sad, as I was quite enamored with my JFET buffer and auto-zeroing scheme, lol.

The evolved circuit, now in its third iteration, appears below. I've shifted the AC/DC coupling relay and cap to after the input attenuator; I decided that I didn't want the additional path inductance between the probe and the shunt termination capacitance. From experience I'm still a bit paranoid about terminating the input of such a high speed op-amp to such a high source impedance. The 1.8pF cap and the 100R resistor in the schematic will be put right at the non-inverting input pin. 1.8pF at 400MHz = 220 ohms, which should nicely keep the input terminated into a low Z at frequencies that could otherwise prove troublesome for pick-up and regeneration. The 100R will neuter any series resonant peaking between the input C and the net trace/component inductance of the loop closed by the shunt termination capacitance over at the input attenuator. Well that's the theory anyway. Proof will come in solder. I'm starting the board layout for a prototype channel now. 

[GK]

Oh, I somehow missed that one. It didn't come up for me in the TI site parametric search.

[GK]

Hmm.... added a SOT23-5 OPA653 to my next outgoing order list. The small signal bandwidth is in excess of my requirements, but the much better large signal bandwidth would be desirable. I'd do two prototype/test layouts; one for the OPA and one for the ADA.
 

[joeqsmith]

Maybe you could roll all three.   Hate to think I spoiled all your fun. 

Hmm.... added a SOT23-5 OPA653 to my next outgoing order list. The small signal bandwidth is in excess of my requirements, but the much better large signal bandwidth would be desirable. I'd do two prototype/test layouts; one for the OPA and one for the ADA.
 

[GK]

That would be nice if there were 48 hours in a day, but I have too many projects on the go and the goal here is just to get a 4-ch probe buffer amp completed. I'll start with the OPA653 and if that works out that is what I'll stick with. 4 channels will be layed out on a single PCB along with a PCB-mount torodial mains transformer and associated power supply. The whole thing will be housed in a die-cast aluminum enclosure.

I wish the '53 datasheet contained a bit of discussion on the thermal characteristics of the SOT-23 package. The only thermal spec. given is a figure of 105 degrees C per W for the thermal resistance junction to ambient. Although that is a useless spec. for a tiny SM device that will dissipate most of its power via the copper traces emanating from its pins if the manufacturers test jig/PCB is unknown it generally is a useful figure for comparison between different devices. 105 deg.C/W seems a remarkably low figure. Having just perused several data sheets for other op-amps in the 5-lead SOT-23 package, I get figures ranging 210 to 280 deg.C/W. I wonder what would make the '53s die/encapsulation such a standout performer in this regard?

The '53 has a quiescent current of 32mA, which is 320mW on +/-5V rails just sitting there doing nothing. For a SOT-23 package with a more ordinary/typical specified junction-ambient thermal resistance, that's getting pretty damn toasty.

Given such a high quiescent current I wonder if the super slew rate spec was achieved solely by running really hot class A bias on the gain stages that drive the compensation capacitance, or if to some degree a "current on demand" architecture is used, which would see the no-load quiescent current and internal dissipation rise further when driven with a signal that exercises the devices full power bandwidth. 

I'm going to torture test the first one I get soldered. In any case it's a given that a decent amount of top-layer copper will need to terminate to the power supply pins and the output pin to get the heat out of the SOT-23 package.


     
     

[GK]

Unfortunately my parts order didn't turn up before the weekend, so I won't have any bench tests to report until next weekend. Here is the layout section for a single channel. Might get the board etched tomorrow.

[joeqsmith]

Looks like your making some good progress with it.    Assuming you ditched the micro.  What's your plan for controlling the relays? 

[Mechatrommer]

so you've ditched out all the mid section discrete jfet input section from the original design and went to raw jfet opamp? thats interesting , i remember deleting my suggestion due to maybe there are specialty in using discrete jfet for a oscilloscope like every brands did. whats more interesting you also ditched out the null offset circuitry, instead of going from mcu to full analog, you go for none at all. i wonder what will be the end output dc offset...

[GK]

Looks like your making some good progress with it.    Assuming you ditched the micro.  What's your plan for controlling the relays?

I have a bag of miniature PCB-mount toggle switches. These will supported on a separate PCB mounted directly behind the lid of the diecast aluminum box that this whole thing is going to be enclosed in. This switch board will connect to the analogue board via a short length of ribbon cable.

I've done some revisions to the 4-ch PCB layout, re-arranging the compensation and termination capacitors of the passive input stage to reduce the length of the inductive tracks between them. I've used 1206 for ease of assembly (home-etched PCB) but if I wanted to go a level better it would be 0603 with components on both sides of the board and lots more (smaller) vias.

[GK]

i wonder what will be the end output dc offset...

OPA653 typical (25 deg C) input bias current and input offset voltage are 10pA and 1mV respectively.

[GK]

One last revision. I decided to add an output offset voltage trim. The PCB layout is done; I'll etch it this evening and load it next weekend. The signal outputs will be BNC panel-mount jacks connected to the PCB via short lengths of coax.

[tggzzz]

That silly squared background on your schematic obscures the content. How? It makes it unnecessarily difficult to see where connections aren't. And that rather negates the whole purpose of a schematic!

While it maybe helpful to you when drawing the schematic, could you please consider your readers by turning it off when publisihing your schematic.

[tautech]

That silly squared background on your schematic obscures the content. How? It makes it unnecessarily difficult to see where connections aren't. And that rather negates the whole purpose of a schematic!

While it maybe helpful to you when drawing the schematic, could you please consider your readers by turning it off when publisihing your schematic.

It's just the visible grid, quite invisible if you are used to it. 
What's more important when laying out schematics and PCB's is the snap grid which may be set the same as the visible grid and in PCB layout packages should be selectable or better still totally user definable.
Everyone has their own preferences. 

[joeqsmith]

Looks like you have given up on the Analog Devices part now and going for a final solution with the TI.   

[GK]

Yes, as I wrote in reply #35.

[joeqsmith]

I feel I need to ask to make sure I get the latest info.      What's you plans for evaluating its performance?   Video in the works?

Would have liked to see if your discrete design would have less noise.

Yes, as I wrote in reply #35.

[Mechatrommer]

What's you plans for evaluating its performance?

it only can be evaluated when the proper pcb is ready and populated, this is a costy and wasty business, from a hobbiests POV. and where experience dominates the theory textbooks.

[Yansi]

Maybe a silly idea, but what about the "classic" solution, using the discrete buffer for AC highspeed signal + DC servo for the rest? (Instead of using veeeeery very fast opamps with fbomb cost level) 

If already mentioned, sorry, I haven't read everything carefully. But it is  definitely interesting topic, as I am currently in the process of designing the same thing (1M to ~50R buffer), but only for <100MHz BW (building it just for my analog circuit curiosity and experimenting) 

[GK]

Would have liked to see if your discrete design would have less noise.

The majority of noise comes from the op-amps and the single OPA653 circuit is less noisy than the two op-amp circuit I followed the discrete JFET buffer with.

Quick analysis:

On the x1 attenuator setting (20dB attenuator switched out). The input is terminated in 1M || 18pF. Consider that the input is disconnected/floating.

1M corners with 18pF at 8842Hz. This is a single pole, low-pass response. A single pole LPF has an equivalent noise bandwidth of pi/2*fo, so:

pi/2 * 8842 = 13890 Hz.

1M generates 129nV/rt Hz of thermal noise at room temperature. The OPA653 has 1.8fA/rtHz input current noise:

1M * 1.8fA = 1.8nV.

Additionally, as noise sources sum as the square root of the sum of the squares, the current noise contribution is totally negligible and can be neglected.

So, the net rms noise contribution from the passive input network is therefore:

SQRT(13890Hz) * 129nV = 15.2uV rms.

This drops even lower when the 20dB attenuator is switched in as the 10:1 attenuator an impedance reduction as presented to the op-amp input.

It's often blithely stated that the noise performance of a CRO/DSO with 1M inputs is limited by the impedance of said input. Until someone can design a 1M input with next to no shunt input capacitance this will remain total BS. It's the wideband amplifiers/ADCs that are the noise performance barrier.

The OPA653 is speced at 6.1nV. Its 500MHz (lets just assume first order for a close enough simplification) frequency response returns a noise bandwidth of:

500MHz * pi/2 = 785MHz, so...

SQRT(785M) * 6.1nV = 171uVrms

Lets sum this with the worst case (15.2uV) noise contributed by the passive input stage:

SQRT((15.2*15.2)+(171*171)) = 171.67uV.

So the 1M input network (worse case) contributes:

log20*(171.67/171)

= 0.034dB to the total rms noise!


171.67uV of rms noise in a 500MHz bandwidth is pretty negligible when viewed on an oscilloscope with 10mV/div vertical sensitivity, so this buffer amplifier won't be a performance barrier in terms of noise performance in almost all cases of use.


NOTE: for simplification I've neglected in the above the <100kHz 1/f noise of the OPA653, but if you do the sums you'll find the contribution to the total rms noise quite negligible.

[GK]

Maybe a silly idea, but what about the "classic" solution, using the discrete buffer for AC highspeed signal + DC servo for the rest? (Instead of using veeeeery very fast opamps with fbomb cost level) 

Deliberated over already. OPA653 is ~$4USD.

[Yansi]

I think ur wrong. 150ns settling time is useless for a 500MHz scope. Or have I missed something? Also the output slew rate will be your enemy, depending on the required output swing. You wont be able to do even 200MHz sinewave at 1V amplitude, if I calculate right.

[GK]

I think ur wrong. 150ns settling time is useless for a 500MHz scope. Or have I missed something? Also the output slew rate will be your enemy, depending on the required output swing. You wont be able to do even 200MHz sinewave at 1V amplitude, if I calculate right.

What? OPA653 settling time (to 1%) for a 4V output step is less than 8nS. Slew rate is 2675V/us. The large signal bandwidth (for 2Vpp output) is 475MHz.




The second sentence in my opening post: "I'm aiming for a bandwidth of at least 250 MHz. While this is only half the bandwidth of the 7904 mainframe I feel that there is no point going any more exotic as the achievable bandwidth with 10:1 10M probes limited."

But to clarify, yes, technically there are 10M probes with >250MHz bandwidth (the 500MHz Lecroy PP006A that I use at work for example) but they won't do anywhere near that without a ship load of overshoot and ringing of their own using the earth clip lead. For that you need to detach the the probe claw and ground clip lead and slip one of those fiddly ground contact springs (with a ~5mm tail that you solder directly to the ground of your circuit being probed) over the collar of the exposed contact tip.

[tggzzz]

But to clarify, yes, technically there are 10M probes with >250MHz bandwidth (the 500MHz Lecroy PP006A that I use at work for example) but they won't do anywhere near that without a ship load of overshoot and ringing of their own using the earth clip lead. For that you need to detach the the probe claw and ground clip lead and slip one of those fiddly ground contact springs (with a ~5mm tail that you solder directly to the ground of your circuit being probed) over the collar of the exposed contact tip.

The lower the tip capacitance the better, of course. There are halfway houses; see the HP probe tips shown in the pictures halfway down https://entertaininghacks.wordpress.com/2015/04/23/scope-probe-accessory-improves-signal-fidelity/

[Yansi]

I might then have downloaded a datasheet for another device.  Seems right now.

[David Hess]

I have considered doing the same thing for my Tektronix 7904 but I only have the 7A24 vertical amplifiers which limit bandwidth to 350 MHz.  That is only 50 MHz faster than high impedance passive probes on my fastest oscilloscope which supports them directly.  If I did do something like this, I would first just build a x10 FET probe to take advantage of lower input capacitance.

For an alternative way to implement this without the performance limiting x2 gain stage, check out the Tektronix P6202A 500 MHz x10 FET probe design.

Tektronix kept the source termination to drive the 50 ohm oscilloscope input but instead of a gain stage, they used a fixed x5 input attenuator so the x2 attenuation of the source termination yields x10 at the oscilloscope input.  The fixed input attenuator has the benefit of making the probe input more rugged under all conditions and it attenuates the input capacitance although they still did not add input protection.

Their P6201 900 MHz FET probe design works more like what you are trying with a x2 gain stage to make up for source termination but they used a discrete two transistor differential transconductance amplifier.  Burr-Brown made some fast integrated operational transconductance amplifiers which of course now Texas Instruments produces and I wonder how well the OPA860 without its output buffer would work in this application.  I remember when these came out and I got the feeling that Burr-Brown intended them to become another basic building block like the operational amplifier but it never happened.

[GK]

Tektronix kept the source termination to drive the 50 ohm oscilloscope input but instead of a gain stage, they used a fixed x5 input attenuator so the x2 attenuation of the source termination yields x10 at the oscilloscope input.  The fixed input attenuator has the benefit of making the probe input more rugged under all conditions and it attenuates the input capacitance although they still did not add input protection.

The downside of which is 20dB less sensitivity - 1V/div minimum as the high bandwidth vertical plug-ins are limited to 10mV/div. Since I have a gain of 2 and can switch out the attenuator I get either 1V/div or 100mV/div. This is only 160mVpp output from the OPA653 for full scale deflection, which is far from exercising the op-amps full power bandwidth.

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

Made a start on loading the PCB. Am only loading one channel to start with for initial testing. The so called SOT23-5 package of the OPA653 is actually quite beefy compared to an ordinary 3-lead SOT23. I guess that goes some way to explaining the much better thermal performance of this package. All caps are NPO/COG.

It's bed time now. With any luck I'll get this channel completed and powered up tomorrow evening.   

[Yansi]

Very nice build!  Please keep us informed 

[David Hess]

The downside of which is 20dB less sensitivity - 1V/div minimum as the high bandwidth vertical plug-ins are limited to 10mV/div. Since I have a gain of 2 and can switch out the attenuator I get either 1V/div or 100mV/div. This is only 160mVpp output from the OPA653 for full scale deflection, which is far from exercising the op-amps full power bandwidth.

You did not say but talked about 4 channels so I assumed you were using 7A24s which go down to 5 mV/div but of course you are right and it would limit your sensitivity even then to 500mV/div.  Are the extra channels for external triggering?

There *is* a way to get 4 channels of 300 MHz performance from standard passive x10 high-Z probes out of a 7904 with probe tip sensitivities from 200 mV/div to 5 V/div without building anything although it is not very common: the double wide 7A42 logic trigger amplifier plug-in.

[GK]

My 7904 is currently equipped with a 7A19 and a 7A24, giving me three channels:

https://www.eevblog.com/forum/projects/old-school-tek-porn/msg419791/#msg419791

Yes you are right, the 7A24 goes down to 5mV; admittedly I have not used it very much so far, lol. My intention is to collect multiple plug-ins for this scope, so I want a four channel adaptor to make use of, say, a pair or 7A24s.

Those low-tip-c Tek FET probes you mention put the source follower (and input attenuator in the case of the P6202) right in the probe tip. This is a rather different thing from an adaptor intended to permit generic 10:1 10M passive probes to be used with a 50 ohm input.

[GK]

Well it seems to work quite well. Now comes the performance testing. This is a bit of an issue because I only have signal generators that go to 200 MHz and none of my squarewave/pulse generators have particularly fast rise/fall times (well not sub nS anyway).



I rummaged through my parts bins and came up with a 10MHz crystal oscillator module and a 74AC14 hex Schmitt trigger, so I made a real quick and dirty fast rise/fall squarewave source. A 74AC gate will do ~1nS rise and fall with no appreciable capacitive load. So I used the 'AC14 to buffer the oscillator module and decoupled its output from any load capacitance with a 10:1 attenuator. The schematic is shown in the picture below. The attenuator gives me a square wave of 500mVpp amplitude with a ~20 ohm source resistance. I am testing with a Lecroy 500MHz PP006A probe which has 12pF of tip capacitance. 20 ohms corners with 12pF at approx 650MHz. Such high speed logic in a big DIP package is never going to make a nice and clean squarewave with no overshoot, as can be seen in the Rigol scope shot (this is not using the probe adaptor).



Here is what it looks like on my 7904 (on the 500MHz effective BW 7A19 channel) via the PP006A 500MHz bandwidth probe and my 1M probe adapter. The measured rise time on screen is 1nS.  BW=0.35/RT, so it would appear that my board layout for the 1M adapter has achieved a bandwidth of at least 350MHz. How much better I of course cannot measure without a faster rise-time signal source and a higher bandwidth oscilloscope and probe.
 

[GK]

Very nice build!  Please keep us informed 

And it seems to work very well  It will eventually get a write up on my webpage. If anyone is interested I can post up the full schematic and Gerber files before then.

[David Hess]

Those low-tip-c Tek FET probes you mention put the source follower (and input attenuator in the case of the P6202) right in the probe tip. This is a rather different thing from an adaptor intended to permit generic 10:1 10M passive probes to be used with a 50 ohm input.

I just used them as examples of what I was describing.  They were optimized for low input capacitance which is why they lack input protection so even with a 1 megohm input resistance, they would not work anyway unless a lot of shunt capacitance was added.

Incidently with a JFET input, usually only 1 input protection diode is needed to prevent reverse biasing the JFET gate.

I rummaged through my parts bins and came up with a 10MHz crystal oscillator module and a 74AC14 hex Schmitt trigger, so I made a real quick and dirty fast rise/fall squarewave source. A 74AC gate will do ~1nS rise and fall with no appreciable capacitive load. So I used the 'AC14 to buffer the oscillator module and decoupled its output from any load capacitance with a 10:1 attenuator. The schematic is shown in the picture below. The attenuator gives me a square wave of 500mVpp amplitude with a ~20 ohm source resistance. I am testing with a Lecroy 500MHz PP006A probe which has 12pF of tip capacitance. 20 ohms corners with 12pF at approx 650MHz. Such high speed logic in a big DIP package is never going to make a nice and clean squarewave with no overshoot, as can be seen in the Rigol scope shot (this is not using the probe adaptor).

The 74AC14 can do a lot better by flattening it to the ground plane and minimizing lead lengths.  Another thing worth trying is using a small signal schottky diode or NPN base-emitter diode between the 74AC14 output and a 50 ohm parallel termination; then the 74AC14 output disconnects itself on one edge from the output circuit and the highest frequency part of the circuit is minimized.

[GK]

The 74AC14 can do a lot better by flattening it to the ground plane and minimizing lead lengths.  Another thing worth trying is using a small signal schottky diode or NPN base-emitter diode between the 74AC14 output and a 50 ohm parallel termination; then the 74AC14 output disconnects itself on one edge from the output circuit and the highest frequency part of the circuit is minimized.

Yeah the construction could be better (like I said, this was quick and dirty), but with the DIP package not by a whole lot. I was contemplating just what you describe, but with a pair of schottky diodes in anti-parallel and the output of the 74AC14 capacitively coupled to the current limiting resistor to give a +/- squarewave with both polarities clamped. Unfortunately the only suitable low capacitance, very high speed diodes I have immediately at hand can't handle any appreciable current.

[David Hess]

The 74AC14 can do a lot better by flattening it to the ground plane and minimizing lead lengths.  Another thing worth trying is using a small signal schottky diode or NPN base-emitter diode between the 74AC14 output and a 50 ohm parallel termination; then the 74AC14 output disconnects itself on one edge from the output circuit and the highest frequency part of the circuit is minimized.

Yeah the construction could be better (like I said, this was quick and dirty), but with the DIP package not by a whole lot. I was contemplating just what you describe, but with a pair of schottky diodes in anti-parallel and the output of the 74AC14 capacitively coupled to the current limiting resistor to give a +/- squarewave with both polarities clamped. Unfortunately the only suitable low capacitance, very high speed diodes I have immediately at hand can't handle any appreciable current.

The idea is not to clamp the output but to disconnect the driver from the transmission line so that any irregularities in its output are attenuated; the output edge is then determined by the shunt termination and reverse biased diode capacitance which can be very small.  The low capacitance of a diode works well for this and many reference level pulse generators work this way.  Of course only one of the edges is the clean one but usually you only need one for testing.  Through hole discrete implementations can easily achieve 600 picoseconds like this (not fast enough for a 400 MHz test) although I would like to try it using faster transistors and a microwave schottky diode from Avago.

The base-emitter junction of a 2N3904 should work well enough if you do not have any small signal schottky diodes.

Anyway, the above is how I handled transient response calibration before I got a sampling oscilloscope which could be used to verify the performance of a pulse generator.

[BravoV]

If anyone is interested I can post up the full schematic and Gerber files before then.

Please do, thank you for sharing. 

[GK]

If anyone is interested I can post up the full schematic and Gerber files before then.

Please do, thank you for sharing. 

Actually it still needs some work. With the attenuator switched in the frequency response is flat as far as I can measure (to ~200MHz) but I have a high-Q series resonance in the vicinity of 500-600MHz that totally screws up the transient response to fast slewing input signals. This seems to be solvable with the inclusion of some small value series damping resistors but this is going to require a bit of work to get right. I am currently working on a spice simulation of my board layout including track and component parasitic inductances as best as I can estimate them.
   

[GK]

Well this was an interesting experiment. I've spent the whole day at the bench, measuring, modelling and modifying this prototype. In the end I did manage dampen the resonances in the attenuator section to an acceptable level, but only after limiting the bandwidth to 250MHz and eliminating the relays from the signal path. I assessed the performance of the attenuator by comparing the transient response to that of a second OPA653 channel assembled without an input attenuator - ideally, the only difference between the two should be in the signal amplitude, not the transient overshoot or ringing. After that I solder wicked all of the remaining input network components from the board and assembled a passive 10:1 (no relays!) attenuator dead bug style as compact as practical right at the input of the OPA653, with the input BNC relocated to the same location. That worked just as well as the unattenuated reference channel in the full 500MHz bandwidth.

In a nutshell, my initial prototype design would have made an excellent front-end for a 100MHz oscilloscope where all of the =>500MHz infelicities would be masked, but as a front-end for my Tek 7904A, it turned out to be a design failure. So now I find myself ruminating over which alternative direction to take. I still want the 20dB attenuator option for the signal handling capability it provides (think probing the high voltage switching spikes produced by a SMPS), but I can't do without the sensitivity of a 1:1 throughput for small signal stuff either. A selection of AC/DC coupling is also mandatory, but the switching between the two at these frequencies is problematic.

So what I am contemplating doing now is making an eight independent channel design. Four of these channels will have 20dB fixed attenuation and four will have no attenuation. Two of the -20dB channels and two of the 1:1 channels will have fixed DC coupling while the other two channels in each group will have fixed AC coupling. Not quite as convenient as only four channels that can have their mode of operation switched, but far better from an electrical, performance and design perspective as all of the input stage switching (relays) and attendant layout compromises are eliminated. I'll probably house these in a 1U-height 19" rack case.

In addition to all of the above I can report happily that the OPA653 is a fantastic performer. In my entire ~8 hours of experimenting at the bench not once did I witness even the briefest burst of RF oscillation.

[GK]

Looking good.

Yeah, the OPA653 performs well, but my initial prototype passive input stage attenuator not so much. See my post above.

[BravoV]

In a nutshell, my initial prototype design would have made an excellent front-end for a 100MHz oscilloscope ..

That is more than enough for me ... preparing popcorn. 

[joeqsmith]

You know, a 1M scope would solve your problem.    Let me know if I can be of more help.   

[David Hess]

For 200+ MHz designs, Tektronix did not use relays to switch their high impedance attenuators until they were part of a hybrid with one exception and in both cases, they used a slightly different topology.  Instead of using a relay to bypass each attenuator, they used it to short the upper section and disconnect the lower section from ground.

Take a look at how the 7A11 does it using a single SPDT relay for each section.  That may return better results.  The hybrids in the 300+ MHz 2465 series use DPDT relays in a similar way.

[GK]

Let me know if I can be of more help.   

You have a spare Lecroy to donate? 


In hindsight with a bit of modelling the HF limitations of the original design are obvious. There are 2 series resonances in the attenuator that were causing me grief. #1 is the the lower leg compensation capacitor resonating with the net loop inductance. This causes a dip in the frequency response that limits the bandwidth. There isn't much that can be done to eliminate this resonance without screwing up the pass band response even worse. #2 is that of the ~9x lower value compensation capacitor in the upper leg of the attenuator resonating with the net loop inductance. This causes a high-Q, high amplitude, high frequency peaking that rings like a banshee in response to a fast input transient if not damped. There is also a 3rd resonance between the input C of the opamp and the track inductance leading up to it, but I had anticipated this resonance and successfully dealt it with the 100R resistor directly in series with the op-amp input.

#2 is easily damped out of existence with a resistor put in series with the upper attenuator leg (but after the input termination capacitance), but with the penalty of bandwidth reduction as this resistor forms a low pass filter with the input capacitance of the attenuator, which in my initial design was ~6pF. Basically, I couldn't snub this resonance (which was at around 500MHz) out of existence without lowering the bandwidth to ~250MHz or less.

The way to deal with these resonances is two fold. The first is obvious and is to simply minimizing all track lengths and associated inductances. The second is to frequency compensate the attenuator with capacitors that are as small in capacitance value as electrically viable and practical. This shifts all of the pesky resonances to higher up in frequency - preferably so high that they either don't matter or so high that they can be successfully damped out without limiting the bandwidth to an unacceptable level. 

Pictured below is a SPICE simulation of the successfully performing  10:1 attenuator mentioned in my second to last post that I built dead-bug style right at the input of the OPA653. Inductance values for component+track/lead contributions are included. The first sim is without the damping resistors to show the high frequency resonances; the second is with the damping resistances added. I think this is a rough estimate of the limit of performance with 1206 passives. Note that the biggest bugbear is the prominent (bandwidth-limiting) resonance of the 15pF cap causing the dip in the frequency response. If this was only a 5:1 attenuator, this capacitor would only need to be ~half the value, shifting the resonant dip up ~two times higher in frequency. I'm not really comfortable using less than 2pF for the (trimmer) compensation cap in the upper leg, so I think for the -20dB amplifiers I will use 0603 passives for the attenuator. This should net me an attenuator that does not substantially subtract from the performance of the OPA653.

I may have "wasted" a bit of time on an unsuccessful prototype, but now I know for sure how to make a s^%t hot 10:1 frequency-compensated attenuator 

[GK]

For 200+ MHz designs, Tektronix did not use relays to switch their high impedance attenuators until they were part of a hybrid with one exception and in both cases, they used a slightly different topology.  Instead of using a relay to bypass each attenuator, they used it to short the upper section and disconnect the lower section from ground.

Rigol gets away with it to 300MHz (I have the 200MHz version). After some more rumination I think I might have been a little quick to condemn the original design. One thing I didn't try on the original layout before stripping it was to simply reduce the value of the attenuator compensation capacitors to one third of the original design values. As detailed in my previous post that would have shifted the pesky resonances to ~3 times higher in frequency and I could have dealt with the small observed irregularities attributed to the relays simply by increasing the value of the damping resistor in series with the op-amp input to lower the attendant low-pass pole to 300MHz. I think all the original design would require for perfectly acceptable 300MHz performance is a small layout modification to include the additional damping resistor in the attenuator network and the above described changes to a few component values.

I could simply revise the original design along these lines...............hmmmm.............. but now that the bandwidth bug as bit..............

[tggzzz]

For 200+ MHz designs, Tektronix did not use relays to switch their high impedance attenuators until they were part of a hybrid with one exception and in both cases, they used a slightly different topology.  Instead of using a relay to bypass each attenuator, they used it to short the upper section and disconnect the lower section from ground.

This is the front end of a Tek 485, a 50ohm/1Mohm 350MHz scope with useful response to >500MHz.

The BNC input, J1, is at the lower left. To its right are three horizontal sections, the 50ohm attenuator chain, the rotary switch positions, and the high impedance chain.

K1S1 is a relay used to disconnect the internal 50ohm termination (so the input is 1Mohm) when there's an overload.

The switches are exposed flexible fingers on the PCB which are lifted 0.5mm by cams on the rotary switch.

The complete manual is available at various places around the web.

[GK]

A few of the 64Xi's relays.   

Very pretty. What model is that from and what is the bandwidth?

[David Hess]

For 200+ MHz designs, Tektronix did not use relays to switch their high impedance attenuators until they were part of a hybrid with one exception and in both cases, they used a slightly different topology.  Instead of using a relay to bypass each attenuator, they used it to short the upper section and disconnect the lower section from ground.

This is the front end of a Tek 485, a 50ohm/1Mohm 350MHz scope with useful response to >500MHz.

...

The 485 was the first design I checked but it uses the custom cam switches Tektronix was so fond of instead of relays and there is no detail in the schematic about the hybrid attenuators although there may be in a different service manual.

Tektronix was making their own relays (and selling them to others) at this time but I could not find any examples where they used them above 100 MHz except at low impedances; the optional X-Y horizontal phase compensation in the 400 MHz and 500 MHz 7000 mainframes used them.  The relays were expensive and so were replaced by cam switches early in the 7000 series products up to at least 400 MHz and later improved cam switches were good to 1 GHz.  I do not know which cam switches the 600 MHz 7A19 used; they may have been improved from the original 400 MHz design.  The two examples I gave used different higher performance relays.

It sounds like GK found the problem though and it was not the relays.  We had a discussion over on the Yahoo Tektronix list not long ago about sweeping oscilloscope inputs to look for high frequency resonances but I did not expect such a good example of the problem.

[tggzzz]

For 200+ MHz designs, Tektronix did not use relays to switch their high impedance attenuators until they were part of a hybrid with one exception and in both cases, they used a slightly different topology.  Instead of using a relay to bypass each attenuator, they used it to short the upper section and disconnect the lower section from ground.

This is the front end of a Tek 485, a 50ohm/1Mohm 350MHz scope with useful response to >500MHz.

...

The 485 was the first design I checked but it uses the custom cam switches Tektronix was so fond of instead of relays and there is no detail in the schematic about the hybrid attenuators although there may be in a different service manual.

Tektronix was making their own relays (and selling them to others) at this time but I could not find any examples where they used them above 100 MHz except at low impedances; the optional X-Y horizontal phase compensation in the 400 MHz and 500 MHz 7000 mainframes used them.  The relays were expensive and so were replaced by cam switches early in the 7000 series products up to at least 400 MHz and later improved cam switches were good to 1 GHz.  I do not know which cam switches the 600 MHz 7A19 used; they may have been improved from the original 400 MHz design.  The two examples I gave used different higher performance relays.

It sounds like GK found the problem though and it was not the relays.  We had a discussion over on the Yahoo Tektronix list not long ago about sweeping oscilloscope inputs to look for high frequency resonances but I did not expect such a good example of the problem.

Good pedagogical examples are worth remembering.

On the principle that some performance improvements have been made in the past 46 years, searching Mouser for "RF Relays" leads to http://www.mouser.co.uk/Electromechanical/Relays/High-Frequency-RF-Relays/_/N-5g33/ That contains 1059 items with frequencies up to 8GHz. Looking at the datasheet for one of those indicates pulse risetimes  <50ps, VSWR <1.1 f<1GHz, VSWR <1.05 f<0.5GHz.

There may be some standard cost-performance engineering tradeoffs to be made

[joeqsmith]

The 64Xi has a bandwidth of 600 MHz.   So many relays, yet so fast.     I've never looked in the 8500A but with the DC coupled, 50 ohm inputs, I would guess the relay count is much less.   If you can't tell, these are a OMRON G6KU-2F-Y.     

A few of the 64Xi's relays.   

Very pretty. What model is that from and what is the bandwidth?

[David Hess]

Good pedagogical examples are worth remembering.

Check off your $100 word for this week.

On the principle that some performance improvements have been made in the past 46 years,

The way I understand it, Tektronix started making these relays because nothing of similar performance was available at the time and they actually ended up OEMing them for others.  The claim to fame for the relays was how they were constructed with the switch sections immediately adjacent to the base for minimum parasitics; modern high performance small signal "telecom" relays use the same configuration and have similar performance.

Where Tektronix could, I think they replaced them simply for cost reasons either with their high performance cam operated switchs or by altering the circuit designs.  Early 7000 plug-ins all used the relays but were quickly replaced by newer plug-ins which did not; the best example might be the 7A16 with 7 relays being replace by the 7A16A with no relays.  Similarly the 7A12 was replaced by the 7A18.

In one way however, modern relays are worse.  The Tektronix relays used a clever symmetrical pinout so the DPDT models can be turned 180 degrees and function exactly the same.  Modern relays use an asymmetrical pinout as I discovered when trying to find replacements for the Tektronix ones.

searching Mouser for "RF Relays" leads to http://www.mouser.co.uk/Electromechanical/Relays/High-Frequency-RF-Relays/_/N-5g33/ That contains 1059 items with frequencies up to 8GHz. Looking at the datasheet for one of those indicates pulse risetimes  <50ps, VSWR <1.1 f<1GHz, VSWR <1.05 f<0.5GHz.

There may be some standard cost-performance engineering tradeoffs to be made

That is definitely the case and more.  RF relays are a lot more expensive than "telecom" relays and being designed for a constant impedance environment, may actually work worse in something like a high impedance attenuator.  I suspect miniature RF relays as we know them were also not available back then either; Tektronix used reed relays with a split coaxial shield instead and they perform well into the multi-GHz range.  There are some good photos of them here:

http://w140.com/tekwiki/wiki/7T11

[tggzzz]

RF relays are a lot more expensive than "telecom" relays and being designed for a constant impedance environment, may actually work worse in something like a high impedance attenuator.

Just so.

I can't imagine using a relay for a high-frequency application without seeing "RF specs" such as VSWR, or at the very least some frequency-dependent specs.

[David Hess]

RF relays are a lot more expensive than "telecom" relays and being designed for a constant impedance environment, may actually work worse in something like a high impedance attenuator.

Just so.

I can't imagine using a relay for a high-frequency application without seeing "RF specs" such as VSWR, or at the very least some frequency-dependent specs.

It would be nice to have detailed specifications but telecom relays are not controlled impedance devices so VSWR and most RF attributes have little meaning.  They sometimes have an ambiguous frequency specification of roughly 1 GHz and depending on the design are good up to at least 500 MHz if the circuit is laid out properly as GK showed.

I used Panasonic TN2-12V relays (since discontinued) to replace a couple of Tektronix relays and they have no RF or parasitic specifications at all but worked great.

[Cerebus]

I used Panasonic TN2-12V relays (since discontinued) to replace a couple of Tektronix relays and they have no RF or parasitic specifications at all but worked great.

If I recall the Panasonic specs correctly, NEC have exact equivalents available. Whether their HF performance is the same is another question.

[tggzzz]

I used Panasonic TN2-12V relays (since discontinued) to replace a couple of Tektronix relays and they have no RF or parasitic specifications at all but worked great.

Many years ago, just as optical fibres were arriving, telecos transmitted 2MB/s down audio grade twisted pair cables specified at 1.6kHz. The telcos couldn't ask for the same cable's specs at different useful frequency, because that would allow the cable companies to increase the price (not cost!).

So they developed test equipment to measure the NEXT and FEXT of the cables, and allow individual pairs to be selected.

[tautech]

I used Panasonic TN2-12V relays (since discontinued) to replace a couple of Tektronix relays and they have no RF or parasitic specifications at all but worked great.

Many years ago, just as optical fibres were arriving, telecos transmitted 2MB/s down audio grade twisted pair cables specified at 1.6kHz. The telcos couldn't ask for the same cable's specs at different useful frequency, because that would allow the cable companies to increase the price (not cost!).

So they developed test equipment to measure the NEXT and FEXT of the cables, and allow individual pairs to be selected.

And in some parts of NZ it's still here. 
Still today telcos try to a sell a shite connection like that to us ^^^^^and have the cheek to call it a fast broadband service. 

[Cerebus]

I used Panasonic TN2-12V relays (since discontinued) to replace a couple of Tektronix relays and they have no RF or parasitic specifications at all but worked great.

Many years ago, just as optical fibres were arriving, telecos transmitted 2MB/s down audio grade twisted pair cables specified at 1.6kHz. The telcos couldn't ask for the same cable's specs at different useful frequency, because that would allow the cable companies to increase the price (not cost!).

So they developed test equipment to measure the NEXT and FEXT of the cables, and allow individual pairs to be selected.

And in some parts of NZ it's still here. 
Still today telcos try to a sell a shite connection like that to us ^^^^^and have the cheek to call it a fast broadband service. 

That 2Mb/s wouldn't have been a consumer connection. It was either a fast (and expensive, think £1000's a month) commercial data link or a voice trunk circuit from a street cabinet multiplexer back to the switch. Back in the day (1995) I used to run a whole ISP on a 2Mb circuit that was split 348 kb/s transit and the balance peering traffic. You could get a lot of dialup users into that 2 Mb.

[tggzzz]

I used Panasonic TN2-12V relays (since discontinued) to replace a couple of Tektronix relays and they have no RF or parasitic specifications at all but worked great.

Many years ago, just as optical fibres were arriving, telecos transmitted 2MB/s down audio grade twisted pair cables specified at 1.6kHz. The telcos couldn't ask for the same cable's specs at different useful frequency, because that would allow the cable companies to increase the price (not cost!).

So they developed test equipment to measure the NEXT and FEXT of the cables, and allow individual pairs to be selected.

And in some parts of NZ it's still here. 
Still today telcos try to a sell a shite connection like that to us ^^^^^and have the cheek to call it a fast broadband service. 

That 2Mb/s wouldn't have been a consumer connection. It was either a fast (and expensive, think £1000's a month) commercial data link or a voice trunk circuit from a street cabinet multiplexer back to the switch. Back in the day (1995) I used to run a whole ISP on a 2Mb circuit that was split 348 kb/s transit and the balance peering traffic. You could get a lot of dialup users into that 2 Mb.

Neither. Remember I wrote "just as optical fibres were arriving". Multimode fibres were just being installed, and single mode fibres were still in the labs.

They were 2048Mb/s links between exchanges, transmitted down some of the quad twisted pairs in a big cable. Only some pairs had the bandwidth, and only some avoided crosstalk.

My second product was an optical attenuation test set.

[David Hess]

I used Panasonic TN2-12V relays (since discontinued) to replace a couple of Tektronix relays and they have no RF or parasitic specifications at all but worked great.

If I recall the Panasonic specs correctly, NEC have exact equivalents available. Whether their HF performance is the same is another question.

I did not find any equivalents but I was not really looking; the Panasonic relays were discontinued after I finished the repairs.  The unique thing about the TN2-12V was that it has the exact same pin spacing as the Tektronix relays and while its pinout is wrong, the coil connections are in the right place and both poles can be combined to replace a SPDT Tektronix relay without changing the physical circuit.

As far as RF performance, these relays all perform about the same if they use the layout with the switching section directly adjacent to the header.

[David Hess]

Many years ago, just as optical fibres were arriving, telecos transmitted 2MB/s down audio grade twisted pair cables specified at 1.6kHz. The telcos couldn't ask for the same cable's specs at different useful frequency, because that would allow the cable companies to increase the price (not cost!).

So they developed test equipment to measure the NEXT and FEXT of the cables, and allow individual pairs to be selected.

This crops up with tested specifications in electronic parts all the time; it is sometimes more economical to do your own grading with your own specialized test than for the manufacturer to do it but you run the risk that your source of "good" parts may dry up.

Usually I see it with leakage in diodes or bias current in operational amplifiers.  Leakage tests at the 50 to 100 nanoamp level are fast making them cheap so that is what small signal transistors are tested at but if you want a low leakage diode, you can grade them yourself and get down to the picoamp range.  The difference between a National LMC6001 and LMC6081 is that the former spent more time (5 dollars worth?) on the automatic tester to guarantee a maximum 25 femtoamp input bias current.

The noise difference between the TL071 and TL081 also came about because of test time although I assume that is no longer the case since they now cost the same; maybe they test them all now and mark TL081s as TL071s as needed.  Low frequency noise by definition costs a lot of test time making it expensive if it has to be tested for.  Popcorn noise used to be the same way until they solved it.

[GK]

No, just busy with other stuff.