Testing unknown transistors for max frequency response

[Argus25]

Originally Posted by G0HZU_JMR

If I try and measure it directly via the 12V 1mA s2p model for the AF178 I get about 0.8 - 0.9pF.

That is bang on, my data book says 0.8pF for the AF178 and 1.5pF for the AF117 or AF127.

It is great you can characterize transistors like this. The lower this value the faster the part at least in grounded emitter stages, in grounded base this miller capacitance is not as important.

[G0HZU_JMR]

I tried analysing the AF117 s2p data at 6V and 1mA and the Ccb capacitance appears to be about 1.3pF. The data is quite noisy again but if I average it I get 1.33pF at a few MHz and the response is flat to quite a few MHz. So this seems quite close to your databook. There will be some uncertainty in these measurements but I have tried to minimise uncertainty as much as possible in my setup. I also effectively measured the AF117 twice with my VNA, once with a mechanical calibration kit and once with the snazzy 13GHz Ecal module and got exactly the same results.

[G0HZU_JMR]

I also looked through the BC556B and BC547B s2p data and these have higher Ccb results. Worse still, these jellybean Si parts show quite a change in Ccb as Vce changes from 3V to 10V at a given collector current. MUCH worse than the Ge parts and it might be worth me comparing a 10.7MHz IF amp design with the AF117 against one with a BC556B. The BC556B will probably show more distortion in a two tone test? But that's just a guess.

[Argus25]

Originally Posted by G0HZU_JMR

I also looked through the BC556B and BC547B s2p data and these have higher Ccb results. Worse still, these jellybean Si parts show quite a change in Ccb as Vce changes from 3V to 10V at a given collector current. MUCH worse than the Ge parts

That is very interesting. My feeling is that the AF178's were one of the peaks of germanium RF transistor development and they are wonderful performers in grounded emitter stages where ideally the c-b capacitance is low.

The world's first commercial transistor radio, the 1954 Regency TR-1 (from the USA) had transistors with such a high c-b capacitance value that they were pretty poor, unless the operating voltage was 22.5V, to reduce the capacitance with the varicap diode effect on the C-B junction's capacitance. So they used this battery, and as the battery went flat the radio would go unstable and the IF's oscillated. And that was with 262 kHz IF's !

However only two years later in 1956 the transistors improved with this parameter and Zenith were making radios running from 6V and with 455kHz IF's too, with transistors like the 2N94. In my post above I tested a 2N94 on my jig and it was not quite as good as an OC45 !

PS: The c-b capacitance of an OC45 is a whopping 10.5 pF which is why in grounded emitter stages they have a limited RF response, but still fine for 455kHz IF stages, if they are properly neutralized. Similar to the OC44 which makes a good local oscillator/mixer for the MW band radio.

[G0HZU_JMR]

I've got an OC44 here. I can look at this one next?

Note that I just spent a bit of time making up a reference design at 10.7MHz. This is all done in the Genesys simulator using the s2p models. To make things simpler for me this uses lumped matching rather than IFTs. I first set the bar at achieving 25dB gain at 10.7MHz with stability factors K>1 and B1>0 across 8-12MHz. It's nicer to have unconditional stability across a wider BW for obvious reasons.

The AF117 is top dog here because it easily optimises to meet the above target. The BC556B fails even at 21dB gain (because its GMAX is only about 22dB!) and the AF178 gets close to the AF117 but can't quite achieve K>1 and B1>0 across 8-12MHz. The AF178 has the highest GMAX but it can't marry high gain with K>1 over a wide bandwidth. The optimiser has the option to damp the LC circuits to trade gain for stability but the AF178 still can't trade its GMAX advantage for unconditional stability over 8-12MHz.

I do all this by allowing a versatile LCR matching network at the input and output of the transistor and I give the optimiser full control over the value of these parts in an automatic optimiser mode. So it hunts down the best solution that meets the above design targets. The AF117 is clearly top dog here. When tested on their own, none of the parts are unconditionally stable across 8-12MHz but the AF117 is the one that achieves unconditional stability the easiest once the LCR parts are added and the optimiser is run.

So they used this battery, and as the battery went flat the radio would go unstable and the IF's oscillated.

I tried swapping across from 10V to 5V to 3V s2p files for each transistor and the AF117 was the one that showed the least change in K factor. There wasn't much in it but the AF117 seems to be the one most at home in my 10.7MHz test circuit.

In the real world it may be the case that all three parts could be used in a real radio and no one could tell them apart and they would all be stable.

I could also try and get hold of some AF127 devices to see if they really are the same inside as the AF117?

[Argus25]

Originally Posted by G0HZU_JMR

The AF178 has the highest GMAX but it can't marry high gain with K>1 over a wide bandwidth.

Is that just the one specimen of the AF178 you have tried ? Because that one seemed a bit odd with the lower Ft than normal and the high current gain ?

I tested all of my AF178's in a circuit at 30MHz (this was of interest as I used them in the EC-10 radio that goes there) with a resonant circuit in the collector and base drive from a generator. In all cases they had a little more output than the AF117 when the collector circuit was peaked. But as I recall the lowest gain AF178 I had was about the same as the AF117's I tried.

[Argus25]

On the topic of Tin whiskers that affect the AF117...I saw a transformer which had a Tin plated bracket that was in a box for probably 50 years, it grew Tin whiskers too, photo attached, its amazing how long those whiskers can get.

[G0HZU_JMR]

I've got another AF178 here I could try. Note that getting higher gain with the AF178 is easy, what isn't so easy is achieving unconditional stability either side of where it is tuned to. The GMAX curve for the AF178 at 1mA 6V is shown below and this shows higher numbers than the AF117. I think it would be even higher if I made a proper fixture for it that allowed short (3mm) component legs.

If the test was just about getting the most gain at the tuned frequency with K>1 then I could just look at the GMAX curve and the AF178 wins. But I made the test harder by making the K>1 bandwidth wider than just the tuned frequency. When I do this the AF117 seems to have an advantage but it isn't a huge advantage. There isn't much in it. If I had set the 10.7MHz gain test to 24dB rather than 25dB and had kept the same K>1 window of 8-12MHz then both parts would have passed the test. The AF117 just scrapes through at 26dB gain and the AF178 scrapes through at 24dB.
If I only set the K window to 10.7MHz +/- a few kHz for the stability requirement then the tables are turned and the AF178 wins out by a dB or so. his is no surprise because this is effectively nudging the GMAX limit at 10.7MHz in each case.

[Argus25]

Jeremy,

Is your test setup basically a grounded base or grounded emitter configuration ? It looks like there is not much to pick between the AF117 & AF178.

[G0HZU_JMR]

Is your test setup basically a grounded base or grounded emitter configuration ?

It's all grounded emitter. All the s2p file data is taken with the device in grounded emitter. There isn't much between the AF178 and AF117 as you say.

If anyone is wondering what the strange knee or bump is in the GMAX curve at 114MHz then this is where the K factor passes through 1.

When K<1 then the device is only conditionally stable. Once it goes above 1 then the device will stay stable no matter what source or load you attach to it. Obviously it can still go unstable if you were to then add feedback between input and output or if there was coupling due to a poor layout. This knee in the GMAX curve always seems to be there when passing through K=1.

You can also see the same knee in my initial attempt at a hybrid model of the AF178 using a VCCS (voltage controlled current source). Even this simple model has the knee in GMAX at 114MHz and this shows that this isn't an issue with my VNA/s2p data for the AF178

See also the plot of the manufacturer's (Infineon) official s parameter data for a BFR106 BJT at 3.5V 7mA. It has the knee in GMAX at 557MHz where K passes through 1. This is their s2p data taken on their VNA setup.

I'm getting a bit nervous that this stuff is getting a bit deep and maybe going off topic? Is it of interest?

[Argus25]

Jeremy,

I think your transistor investigations are amazing and exceed anything I could do.

I guess we have figured out that a transistor like the AF178 is a good replacement for AF117. The question will come up though, which other types might be. As really, all AF11x should be replaced in radios.

There are a lot of unmarked germanium transistors out there, that might or might not be ok, there are lots of Russian types too. So I guess its a matter of cooking up some sort of recommended test circuit to quickly find out.

Hugo.

[G0HZU_JMR]

I think some care would be needed when setting up a test circuit at 10.7MHz because it's possible to get oodles of gain if the circuit is tweaked to the point where it delivers a form of regenerative gain. For example, if I take the brakes off the optimiser (by removing the stability limits on K and B1) and let it freely tune the RLC networks on the input and the output of the BC556B at 10.7MHz I can demand a gain of 45dB at 10.7MHz and quickly get it.

See the simulation below. The BC556B gives about 45dB gain here. K is only just below 1 but B1 has gone insanely negative and the amplifier is more like a regenerative stage in a simple SW receiver. i.e. loads of gain, close to oscillation and lots of 'selectivity'.

If I leave these RLC values alone and drop in the AF117 the peak response shifts up to 12MHz and the gain falls to about 23dB here because the AF117 isn't close to oscillation. K and B1 indicate unconditional stability. If I then let the optimiser run again then I can get >40dB gain at 10.7MHz from the AF117 at the cost of K and B1. However, despite its higher GMAX the AF117 struggles to go into regenerative runaway and it takes the optimiser a long time to get 45dB gain like this.

It had to remove both damping resistors in the RLC network and then optimise the LC values to get it to misbehave like the BC556B did.

So I think a 10.7MHz test circuit would need to have a fair bit of damping in it to give a fair comparison between the transistor types to prevent regenerative gain from masking the true results. Otherwise the BC556B would probably be the winner in terms of getting the most gain in the stage.

[G0HZU_JMR]

The reason both these transistors (and the AF178) can do this 'regenerative' gain trick is because they are all 'conditionally' stable when tested directly for K and B1 at 10.7MHz. So they can be coaxed into a state where they will give regenerative gain just by tweaking the input and output networks for more and more gain. Each one will have its own sweetspot combo of RLC networks at the input and output that kicks it into this mode. In other words, there's no need to add an external feedback path to explore regenerative gain with any of these transistors at 10.7MHz. We've all been there when one keeps tweaking a homebrew IF stage for more and more gain and eventually it gets so much gain it gets hissy and then bursts into oscillation

[G0HZU_JMR]

I tested an OC44N today on the VNA from 0.3-100MHz and the Ft at 10V 1mA was 12.9MHz.

The GMAX gain curve is given below at 10V 1mA. This shows this device is good for just over 30dB of gain at 455kHz if you want to retain unconditional stability.

The K=1 'knee' in GMAX is at 1.8MHz and the GMAX gain drops to about 7dB by 10.7MHz. So not much use as an IF amplifier up at 10.7MHz but I guess everyone knew this anyway

It looks like it can be configured as an oscillator up to just over 20MHz but this is really pushing it. I haven't built an oscillator to prove this as I don't want to risk popping this transistor.

[Argus25]

Originally Posted by G0HZU_JMR

I tested an OC44N today on the VNA from 0.3-100MHz and the Ft at 10V 1mA was 12.9MHz.

That is about right, the book says >7.5 < 15 for the Ft for OC44, and 3 to 12 MHz for the OC45, but the OC45 I tested , I guesstimated the Ft at about 3.5.

[G0HZU_JMR]

I got an AF127 today to play with. This was sold as new old stock. It's definitely an old device and it looks genuine to me. It has the correct (strange) pinouts for ebc and shield.

However, this appears to be a different animal to the old AF127 datasheets I can find. This transistor is quite fast. From the s2p data I get an Ft of 185MHz at 1mA 6V. I'm unlikely to make a significant error with the VNA setup and I was expecting an Ft of 75MHz. So I'm quite confused here.

The feedback capacitance is low as well. 1.3pF instead of the 3.5pF quoted on the datasheet. The K factor doesn't cross 1 until 112MHz when tested on its own. So this is different to all the other devices I've played with. It's definitely a Ge device because of the Vbe voltage.

I guess I'll have to try and find another one to compare it with.

[G0HZU_JMR]

I found a fairly good datasheet for the AF124/5/6/7 online here:

https://www.semicon-data.com/pdf/SH/AF/AF127.pdf

I converted my s parameters to Y parameters in my simulation SW and compared mag[Y21] and my results are the exactly the same as the mag Y21 graphs in the AF12x datasheet up until about 20MHz. But my results don't droop above this frequency like the datasheet. So I don't think I have a genuine AF127 here. It's much closer to an AF178 in terms of frequency response. I double checked everything and even put this device in the tuned 70MHz amplifier test rig I have here and it gave 16.3dB gain at 7V 1mA and my s2p data predicted 16.4dB. The input and output impedances agreed very closely so I don't think I'm making any (significant) measurement errors here. I also got an Ft of 180MHz in my homebrew Ft test jig. Also, the jig showed x2 current gain at 90MHz and x4 gain at 45MHz all of which point towards an Ft of about 180MHz at 1mA and 6V.

I looked at the transistor up close at x25 magnification under a decent stereo microscope and the fonts used for AF127 have a distinctive 'edgy' look to them as if they have been generated on a computer. i.e. the curve on the '2' is made from straight edges in places like it has been cropped into a curve with scissors. This is compared to the soft analogue appearance of all the fonts on all the old metal can transistors I have here that are from the 1980s or older. You can't see this unless a decent magnification is used.

I think it would be wise to try and find a genuine AF127 that has come from a known bit of vintage gear from 40 years or so ago. I should be able to replicate the Y parameter graphs in the datasheets for all of these devices. If my results match the datasheet then I can release some decent/modern s parameter data for these devices with a degree of confidence that I have a real AF127 here.

Ultimately, it should be possible to make some form of simple test jig that can tell all these devices apart.

[Radio Wrangler]

You might be able to concoct an oscillator circuit that either will or won't start as a jig to tell them apart.

There are still some transistors of rather dodgy stability in the modern era. Take a look at the datasheet of MRF1513 and then notice how much is measured with an RC network in series with the gate connection! It's not shown in the schematics, but notes in the text say it's there.

K factor can be a bit misleading. device whose K goes below unity can only be conditionally stable even if you've matched it for and are using it at a different frequency where K is safely over 1. It all just means that if the match you use happens to look unfavourable in the K<1 band, it will hoot at that frequency. Stability has to be achieved at all frequencies simultaneously. I use some little MMIC amps whose gain goes uo like a ski-jump off their LF end AND Zs becomes difficult to keep resistive in that range. A bit of a balancing act.

David

[G0HZU_JMR]

Yes, to try and capture out of band behaviour for K I measured s2p data for the BC547B from 300kHz up to 1GHz. For the Ge parts I've limited this to 500MHz but that is a bit marginal in a couple of cases.

I got another AF127 today. But already I think it's a dodgy example because I looked at it with a microscope. At x25 magnification it's easy to spot that someone has (almost) rubbed away the original markings and painted AF127 on the metal can. The new logo is offset from the old one and with a different font so this is easy to spot.

The original markings begin with AF12 and end with A7 but they are very faded and are partially covered up by the new logo. It might be that this was done at the factory for some reason but it does seem a bit suspicious.
It looks very much like the original logo was AF126 A7 but I can't be certain. Having looked at the datasheet there isn't much between the AF126 and AF127 so I think it's worth testing. The AF127 seems to be the poor relation so maybe a batch of AF126 devices got downgraded at the factory to AF127?

[G0HZU_JMR]

I tested the latest AF127 and I've called it AF127b. This proved to be slightly faster than the previous AF127! The feedback capacitance was just under 1.2pF as opposed to 1.3pF for the first AF127.

Ft at 1mA 6V was 202MHz for this transistor. When I looked at both of them side by side under the microscope and ignored the labels the construction looks to me to be identical. You can see a lot of subtle detail under a decent microscope and the metal package looks to be very similar and so does the material used to seal the area under the transistor body.

To kind of prove I'm not messing up my measurements I put my VNA derived 6V 2mA s2p model of this AF127b into the Genesys simulation of the negative resistance oscillator circuit I showed earlier in the thread for the BC547B. This then predicted that with the AF127b fitted the maximum frequency of oscillation (in this circuit config) will be just over 420MHz before it dies.

So I built the oscillator using the old parts from the BC547B oscillator but obviously I had to swap the PSU voltage and also I had to change the bias network to suit this PNP AF127b so it operated at 6V Vce and 2mA. But it behaved exactly as per the simulation predicted. I started off with a high inductor in the collector and trimmed it down and watched the oscillation frequency go up until the oscillator stopped and wouldn't restart. This happened at 426MHz and the simulator predicts about 420MHz. You can see the graph below shows the resistance ceases to be negative at about 420MHz (red trace) and the simulation includes the ~30nH inductor in the collector so it is showing resonance at 420MHz in the blue trace on the graph.

So the fact that this thing can oscillate up at 426MHz at just 6V and 2mA shows that it is quite a fast transistor. But is it a genuine AF127? The faded (original?) logo looks like AF126 but there is a tiny impact mark over part of the 6 so I can't be certain. You might be able to see the faded logo on the picture of this transistor below.