1MHz output impedance query?

[Radio Wrangler]

Stop trying to save components at all costs!

Resistors are cheap. Especially simulated ones.

Do your base bias with a two resistor potential divider. That one resistor approach is common in cheapskate lousy equipment, but the drop depends on bas current which depends on Hfe current gain which is a very uncertain parameter. It's how people dsign circuits which don't all work very well, and force people to select transistors. A resistor or two extra for rock-solid control of the quiescent conditions is really a bargain.

David

[mole42uk]

That could be a feature of using a simulator - it works based on the simulated component parameters. Real life may be disappointing!

I was holding back on this one, until now. With modern high performance op amps the whole thing can be done with just op amps, resistors and capacitors.

[mole42uk]

MM - yes, I see your point but it's now become a learning exercise for things I should have done years ago. Not only RF filters, but LTspice as well, which I always knew was a useful tool but I never learned to use it. I still have a long way to go.

Here's my latest simulation, biasing the transistor properly and I've decoupled the output of the filter and let it see a 2m piece of 50Ω RG-223 on the end.

[Radio Wrangler]

Now, let's say your emitter follower is a wonderful one. Maybe a few concatenated. The Zout will be roughly re=25 Ohms/IC in mA

That's likely to be rather small. This is actually a problem!

Stand by for a mind-altering concept!

If I gave you a load of perfect inductors and capacitors of all values... could you make a perfect filter?

Well if you wanted an infinitely fast roll-off you'd need to use an infinite number of stages.

But perfect components are lossless.

So where did the energy you stuffed into your perfect filter go if it didn't come out the far end?

There is only one place. Bounced back out the input!

So the throughput and the input reflection have to be complementary. So we have to dissipate it, else it keeps bouncing to and fro until it eventually gets into mischief.

So the interesting bit of your filter, where it is starting to roll-off is critically dependent on the impedance it is fed from. The load impedance is also important near and in the passband.

So you need a resistor between the emitter and the filter.

There is a second reason for this resistor.

Emitter followers are notorious for oscillating due to stray C and L making a sort of internal Colpitts. So the shunt C of the filter is risking it. Good housekeeping by experienced (bitten!) designers sticks a resistor in the base connection or else a ferrite bead to tame the little blighters.

David

After using LTSpice for years I have noticed the "battery" symbol on your simulation, that will make it easier for me (at least) to discriminate twixt signals (a voltage source) and power supplies (another voltage source).

[G0HZU_JMR]

The emitter follower circuit looks a bit odd to me for various reasons. I'm assuming your logic gate has a series 10k resistor after it and this forms a switchable level at the base of the BJT. It looks like the BJT will turn on and off a bit like a switch as the 10k will act as a pulldown at logic 0 and then the logic 1 will turn on the BJT.

However, I'm a bit confused by the circuit in the emitter. it looks to me like there isn't a well defined (as in stiff) operating point because there are 10k (as in large value) resistors in the emitter. I'd expect the circuit to pump up the emitter voltage a bit like a peak detector because of this. There doesn't look to be a way to pull charge out of the capacitors in the emitter circuit in other words. It doesn't look like it will work properly to me as it looks like it will tend towards a peak detector in the steady state.

The output impedance won't be 50R either so this doesn't meet your earlier requirements.

Maybe it would be good to decide what signal amplitude you want for your 1MHz sine wave when driving a 50R load.
It might be good to explore what can be done with a simple HC logic gate driving a classic pi matching network via a series resistor. With a 5V supply and about 200R series resistance you should (in theory) be able to extract 1.25/200 W =0.0057W. This is just over +7dBm into 50R.

If a bit of loss is allowed for in the matching network then this would deliver +7dBm into 50R with a nice sine wave. This will probably only need 5mA (possibly 4mA) from a 5V supply to deliver a +7dBm sine wave into 50R at 1MHz. This is just under 1.5Vpkpk.

If you want more level than this then maybe use an AC gate and 120R series resistance. This will deliver 1.25/120 = 0.010mW which is about +10dBm or 2Vpkpk. This is a fairly large signal and it might be better to place a 3dB attenuator at the 50R output to deliver +7dBm via the AC logic gate. This will give a nice 50R source impedance at 1MHz and the 3dB attenuator will help provide a resistive termination across a wider bandwidth as it will be just after the matching circuit.

[Radio Wrangler]

Driving say a 50 Ohm load to a moderate level without eating a load of power supply current is a common enough problem.

With your LTspice circuit, you'd need a blocking C from the signal source, so your bias network can do its job. Then you need to make sure the emitter bias resistor pulls enough current that the needed negative going swing with your 50 Ohm load attached doesn't take the transistor into cutoff or try to take it beyond.

More sophisticated drivers use a totem pole emitter follower that can pull up or pull down in class B or at least in variable bias class A. If you go this far, you might as well buy a fast opamp which hasa a bit of poke. I used to use AD811 but there are more modern alternatives that are easier to power.

David

[G0HZU_JMR]

As drawn I think the existing emitter follower will be switched off when the gate drives to a logic 0. I think this means the follower isn't a follower anymore and when the gate driver is at logic 1 the BJT will pump bursts of current into the emitter circuit that will charge up the 100uF blocking cap. I suspect that the circuit will start to misbehave as that huge 100uF cap gradually gets charged up. The LTSpice simulation seems to be very brief and it doesn't predict the steady state condition because this huge cap is probably still charging up.

I was expecting to see the matching filter ahead of the emitter follower as this allows the emitter follower to define a very broadband 50R source impedance via a series 47R from the emitter to the output via a dc block.

I can't see the point of the current circuit topology of follower>>filter when compared to the classic logic gate circuit driving a matching filter via a series resistor and a blocking cap. This will be simpler and more efficient. I think the AC logic version can deliver about 2Vpkpk into 50R. I think the power supply current from 5V will be about 7 to 8mA in this case.

In comparison a basic emitter follower will probably draw 25-30mA from 12V to deliver the same 2Vpkpk into 50R if the aim is to have similar distortion performance.

[Radio Wrangler]

Filters don't have to be designed for 50 Ohms, but 50 Ohms is useful for an output driving coax t wherever it gets used.

So why not have a light-ish powered buffer from the logic into a filter designed for higher impedance, then a beefier buffer into the 50 Ohm cable?

Alternatively the first buffer could be a bunch of CMOS gates in parallel driving a 200 Ohm resistor into a 22o Ohm designed filter (allowing for Zout of the paralleled gates) then the filter into a buffer driving the 50 Ohm world?

There are lots of possibiiies. It can be difficult to choose.

David

[G0HZU_JMR]

I'm not too keen on the current emitter follower circuit but maybe it will evolve into something better.

However, can I suggest the classic

74AC>>series_100R_resistor>>dc_block>>pi_match>>ou tput

as a benchmark for a +10dBm 1MHz sinewave generator with 50R source impedance?

Whatever circuit gets developed here it can then be compared to this benchmark circuit that can deliver a 2Vpkpk sinewave into 50 ohms.

The AC gate version should typically draw about 7-8mA from +5V and this is about 25% efficient.

Other things to compare are harmonic rejection and accuracy of power transfer (at 1MHz) into various load impedances up to (say) a VSWR of 10:1 that can get presented at the output. Ideally, the circuit should behave as a theoretical 50 ohm source into all these loads. The circuit should also still deliver a decent sine wave in all cases with low distortion.

The main weakness of the 74AC circuit is that it only has a 50 ohm source impedance close to 1MHz. Usually this won't be a problem but it might be better to put a 3dB attenuator at the output as I mentioned in an earlier post. This will reduce the output to about +7dBm but it will mean the worst case source VSWR will be 3:1 at frequencies well away from 1MHz.

I also agree with CM that it is worth drawing up a list of requirements as this will help to define the design approach better. It looks like this is for a general purpose 1MHz sinewave source with 50 ohm source impedance that could get connected via coaxial cable to various bits of test gear that might or might not have a 50 ohm input impedance. I would suggest that +7dBm into 50R is a good target but maybe +10dBm might be needed in some cases? -40dBc harmonics would be a nice target and maybe 40mA max current from +12V. It should also behave into various loads that are far away from 50 ohms.

[mole42uk]

This idea developed from a "simple" requirement: I am building a GPSDO primarily to provide a known reference for my Marconi 2955 Radio Test set. This wants a 1MHz reference and seems to be quite unconcerned as to the impedance or amplitude of the signal.

Reveiwing the designs that others have produced, I first tried to use James Millers idea and a Jupiter GPS receiver. After a lot of work I found that the GPS receiver was faulty so decided to update the project and use a ublox receiver and a different 10MHz OCXO. This has been very succesful and locks quite happily so I have achieved the first requirement and have the bonus of a solid 10MHz reference as well. The device will remain powered all the time (24/7 as the Americans would say).

Subsequently, I thought that it would be useful to provide a 1MHz general purpose output. As all my current test equipment uses 50Ω connections, I want a 50Ω output and since a low-distortion sine wave would be safer as a general purpose signal, I started to think about square-to-sine filtering. Where I am now is a result of a) not knowing enough, b) discussing it on here with people who know what they are talking about and c) learning some of the mysteries of LTspice.

The circuit I presented above seems to work in LTspice, even if I make the duration much longer until it acheives a steady state but I am happy to acknowledge that there are things I've missed.

So, the specification develops:

1. 1MHz low-distortion sine into 50Ω coax
2. A reasonably high output level - my BBC-biased audio engineer head thinks 0dB but that might be just silly at 1MHz - but a switched attenuator could be used
3. The power supply can offer 12V at 100mA
4. Stability into loads that are not 50Ω

[Radio Wrangler]

0dBm (1 milliwatt) to +5dBm is a common sort of level for standard frequency distribution as a sinewave.

Sinewave distribution means there aren't any harmonics or rings which could give a double bump problem where a ripple on a rising/falling edge gets counted twice, effectively doubling the frequency the recipient sees.

10MHz is perhaps the most common reference frequency that gets distributed to test equipment.

A large lab or factory would buy one precision frequency source, a distribution amplifier and a lot of RG58. All their instruments would be ordered without high stability ovened oscillator options, saving money, and would use the 10MHz distribution system.

David

[G0HZU_JMR]

Could you share your LTSpice .asc project file by attaching it as a zip file? This way it could be analysed by other LTSpice users.

[mole42uk]

Here's the file - at least I think it's the .asc

[G0HZU_JMR]

Thanks! I had a quick play with it and does seem to be behaving better. However, the output level is quite low. Also the pi filter values seem a bit odd and the BJT operating point will need to be changed to suit larger signal levels.

I also edited the circuit values back to the previous version shown in post #24. This earlier version of your follower circuit does seem to (kind of) behave in the way I predicted. In the image below I simulated out to 100ms. This is still only about a blink of an eye after power up but it shows the output level slowly collapsing (blue trace). It also shows the emitter voltage slowly pumping up (green trace) and this is slowly biasing off the BJT as the circuit settles.

At maybe 1ms it shows about 2Vpkpk at the output (blue trace) but this collapses to a small level after 100ms. With a real build of the circuit this might fall even more. I'm not sure how the 2N3904 model will behave in this circuit because there is the pumped up voltage on the emitter trying to push backwards through the B-E junction as the junction spends more and more time reverse biased. The BJT model might not be able to cope with this unusual situation very well although it will probably do a reasonable job.

With all this in mind, one thing to always watch out for with transient simulations up at RF is that it's tempting to only run the simulation for a few microseconds or milliseconds after power up. This is because the simulation will be slow if longer simulation times are selected. Sometimes the circuit won't have settled and what is seen in the graphs won't represent what we would see on a scope after the real circuit has fully settled.

[mole42uk]

Ah, some light dawns - I haven't been leaving the LTspice simulation running long anough.

Pi filter values came from one of the ideas in post #6 and some results from a couple of on-line calculators, modified to 'improve' the output. I have been thinking about driving the pi from a 74HC14 but I can't find, or calculate, the output impedance of the HC14 at 1MHz - that's been awkward!

Today, I think I've been trying to do something way above my pay grade. I'm accustomed to 10kHz, but 1MHz is a very different animal.

[G0HZU_JMR]

I'm not free until this evening but I can take you through the formal design process with the HC (and AC) gates later.

I can give a few snippets straight away though... a typical HC04 gate has an ESR of something like 30-40R but you can't match to this. You have to add a (wasteful) series resistor to the output in order to prevent excessive current being drawn from the HC gate. In the past at work I always used a series blocking cap and a series 270R resistor. This gives an AC source impedance of 300R. You then follow this with a pi match from 300R to 50R at 1MHz.

The short form of the design equation (that I use) is that the power delivered to the matched 50R load will be 1.25/Zsource = 1.25/300R = 0.004W or 4mW if the HC gate is run from a +5V supply.

It's probably OK to use a Zsource of 220R and this would mean using a series 180R resistor (and maybe a 100nF blocking cap) in series with the 40R ESR of the HC gate output. The power would then be 5mW or +7dBm into a 50R load.

You should get a fairly accurate +7dBm and a 50R source impedance at 1MHz with this setup providing you design the matching section correctly and don't use lossy parts.

[G0HZU_JMR]

See the LTSpice screenshot below to get the general idea. I've only used LTspice a handful of times but the plot below shows the power in the 50R load resistor averaged across 0ms to 5ms.

This should show a sinewave response for the power at 1MHz if I zoomed in. This shows a peak at 8.2mW. The average power will be half of this at 4.1mW as the trace is symmetrical about 4.1mW.

Note: This isn't the same as working out the power using Vrms. The simulator has already done this bit for me and it is displaying power (not voltage) so the mean power will be half of 8.2mW and the little text box has calculated the mean power the hard way across all of the waveform across the 0-5ms span. It agrees at 4.1mW mean power.

The design equation predicted 1.25/300R = 0.00417W which is close enough to 4.1mW. This does use a lossless inductor in the matching section though...

[G0HZU_JMR]

Note also that the 50R resistor shown as R5 is the termination resistor in your piece of test gear. This resistor can be at the end of a run of coax for example.