Make a very simple test for me

I think the problem is because you are using silicon transistors which have a higher base emitter conduction voltage (circa 0.7V) than germanium ones (circa 0.2V). The small difference matters when Vcc is only 2V. In the early days germanium was more common than silicon. With a low VB voltage of 2V the transistor BC548 may not be in the required operating zone.

Just to see if the circuit does oscillate replace the 2V battery with say 5V- 6V and tell me what happens and you see on your scope. Also check you have connected the power supply to the circuit correctly so that VB is positive and the ground rail is negative since the transistor is npn type.

Well that oscillator would be for series resonance between the two low-impedance emitter points of the two transistors. However, my immediate reaction to looking at the circuit is there is no attempt to stop DC positive feedback, so it could end up resting with one transistor saturated and the other cut-off. You might need to make the emitter resistors Re larger than the collector resistors Rc to stop that (i.e. so each transistor's DC gain is less than 1) and then rely on the crystal's series impedance being low enough that you have greater than unity AC gain for the oscillations to start.

A typical 20 MHz crystal would have a series resistance of something like 100R or that sort of order, so trying a swap with Rc = 1k and Re = 10k might be happy.

You also need to consider the loading of the scope, etc, on the collector of Tr2 as that will reduce the gain (if AC coupled) and possible impact on DC bias (if DC coupled). You might want to put a capacitor of some 1nF or whatever and a resistor of over 1k in series as a means of tapping off the signal without too much load. Or use a 3rd transistor as an emitter follower buffer (Tr3 collector to supply, Tr3 base to Tr2 collector, Tr3 emitter with 1k to 0V, output from Tr3 emitter via 56R and 1nF in series).

Incidentally you would not believe the number of different crystal oscillator circuits that have been developed over the years!

They span all sorts of different use-cases and part of that comes from the technology for amplification (thermionic valve where one is as much as you want, discrete transistors where over 2 is becoming a crowd, and IC with complexity as cheap), the frequency range needed (so low frequency has high Z crystal so high Z oscillator needed, but very high frequency means forcing resonance on an overtone via frequency-selective amplifier, etc) and the desired purity/stability of the output (typically meaning operating the crystal with a well-defined sine wave drive level and optimising the in-circuit Q factor).

Whole books have been written on the subject, for example:

That simple transistor multivibrator was probably running at a much lower frequency and might not be capable of operation at 20MHz due to lack of loop gain. Quartz watch crystals standardised on 32.768kHz very early on, I don't know what the 'first' one would have been but probably no higher in order to minimise dynamic power consumption.

As above there are many different circuits; I would not choose this one. Something like a pair of logic inverters would seem to offer more predictable results.

E2A A little googling suggests that the crystal in the first version of the 35SQ ran at 8.192kHz

Fantastic explanations ! I love them all.
I will update you with your recommended modifications in a short time.

As I suspected, I actually have that book in hardback:

I did first @marconi suggestion since it was the simplest to make.
And the same result as before. I test it at 2V, at 5V, at 9V (from my VPSU (Variable power supply unit)) and the same output on all 3 = a flat line. Well, a Parkinson flat line.
Here is the prove:
I did the 2V test anyway, just for the sake of comparison. I do believe your idea with the germanium and silicon diferences. And I know it myself but I didnt think on it. On time. You were faster. Which is good.

The fall:

Do you see what I mean by Parkinson "flat" line? And that my DSO138 is dinky as hell?
You don't believe me. But Im telling and showing you.
You have EXCEPTIONAL osciloscopes compared with my dinki one here.
Anyway...
The rise:


Now at 5V

the fall and rise : (less than 1V)


and finally at 9V

the rise and fall:
notice this is spot on 1V - fix in the middle of the 2V per division square.


and my original circuit:

Im showing here how I connected the osc probes. Green link = poz + and yellow link = neg -.
This is still the original circuit.
Sorry for the blured camera but I dont have my other one at the moment.
I will now build and change the other recomandations !

As pointed out by @Lucien Nunes the original watch circuit would use a low frequency crystal.

These days they are all 2^15 = 32,768 Hz for ease of dividing to 1 Hz for counting seconds, etc, and are quite cheap and available in many places. The original choice might have been less as Lucian's points out in the days before dedicated IC dividers and each step was done by using separate parts.

Low frequency crystals are troublesome beasts, they can be large unless 'tuning fork' style, have a high series resistance and today you would struggle to find anything that is not a mass-market frequency. In the past we got custom crystals from the likes of McKnight Crystals in Southampton, England but that sort of small company has long gone as standard frequencies were all that most need due to frequency synthesisers and related microcontroller control, etc, became cheap and ubiquitous.

Your circuit, assuming ideal transistors, would present a negative resistance of around 2*Rc between the emitters (basically a delta-I current through the crystal products delta-I*Rc voltage on both emitters) so if you get a 32kHz watch crystal such as this:

The data sheet has its resistance as 35k Ohm max, however, you need a bit more than that to reliably start up so probably you would be looking at 100k for the Rc values.

To preserve DC stability you need Re to be greater than Rc, so you might be looking at 330k or similar, also to keep operating current down to prolong battery life and to avoid over-driving the crystal (no spec on data sheet for drive level unfortunately). With such high impedances you would need to make sure you don't load the collector too much to extract the signal, so either a buffer or a 100k resistor + coupling capacitor, etc.

Lucian also mentions about the transistor frequency response, that is important as it limits the frequency at which you can get useful gain and (less obviously) it is also current-dependent in the transistor. Looking up the BC548 data-sheet the fT is nominally 300MHz at 5V and 10mA, but could well be below 10MHz in this circuit, so a high frequency crystal may not work without pushing down R to get I up for the transistor.

@marconi also raised an important point about biasing, but with 2V and approx 0.7V (a little less at low currents) for the B-E junction you should be seeing around 0.1mA or so per transistor with Rc = 1k and Re = 10k (so fT probably around 20MHz, marginal here with 20MHz crystal). Hence if you want to use it at 20MHz you might be better to accept high current and not lithium cell powered and just set the supply to 5V and maybe drop emitter R to 2.2k

pc1966 said:

...32,768 Hz ...

Click to expand...

Damn. I made a big mistake. For some unexplicable reason I read the "20 000" marking on my OSX (crystal oscillator) as Mega. While it is 20 Kilo. Damn. And I can not EDIT the original post either to correct it... what a shame. Again, Im not working as often as you guys here, with these OSX's. My VERY simple goal here is to measure some Blank OSX's I have scrapped and collected for a long time. Blank = no markings, no letters or numbers on them. Simple. And quite recently I learned that I need a specialized tool for it, the frequency counter. That I am waiting for at the moment to arrive in couple of months. It would be good to be able to measure the fv only with the osciloscope, and I bet it is possible, but I am very sure not with what I have here.
That is a very nice book, by the way !

Hmmm, I just checked the other known values and I didnt give this much attention to this detail which is very important. Which are the Kilo and which are the Mega. I ASSUMMED "20 000" marking is Mega:
So this is the OSX I used so far in all my tests and circuits:

As you can clearly see it does not have a K or M indicator at the end. Thats why I assumed and now I think probably certainly wrongfully.
Because, I remember the Mega on some of my OSX's and I check it and I find one:
I know is not THAT visible but it is good enough. It says 27.000 MHz.

So a "20 000" without the MHz specified in its end, it might simple be a 20 KiloHz.
Ha.... Right? I am thinking out loud here. Again, I am working with these components almost never. I am not used to them. I am not using them in my circuits that often. All that I used them for was for my PIC MCU external clock signal, and thats this shortcut term I borrowed from as well. Also I didnt used them for PIC's as often as well either. But I thought, to have a bunch of these OSX for some future experiments that never happened.

_q12x_ said:

Damn. I made a big mistake. For some unexplicable reason I read the "20 000" marking on my OSX (crystal oscillator) as Mega. While it is 20 Kilo. Damn. And I can not EDIT the original post either to correct it... what a shame. Again, Im

Click to expand...

A 20kHz crystal would be very unusual, especially in that sort of a package. I suspect 20MHz as you initially thought.

_q12x_ said:

not working as often as you guys here, with these OSX's. My VERY simple goal here is to measure some Blank OSX's I have scrapped and collected for a long time. Blank = no markings, no letters or numbers on them. Simple. And quite recently I learned that I need a specialized tool for it, the frequency counter. That I am waiting for at the moment to arrive in couple of months. It would be good to be able to measure the fv only with the osciloscope, and I bet it is possible, but I am very sure not with what I have here.

Click to expand...

If you can measure the time between successive edges with the scope then freq = 1 / time

However, most scopes are not very accurate at this, even the ones that have built-in capabilities to estimate frequency. You will get 0.1% or so accuracy often (i.e. 1E-3), not the 1ppm or better sort of thing you get from basic frequency counter (i.e. 1E-6) such as:

_q12x_ said:

That is a very nice book, by the way !

Click to expand...

It is handy, can't remember when I bought it but I thing a couple of decades ago. This presentation might be of some interest to you:

Looking up your DSO138 oscilloscope it appears to have a 200 kHz bandwidth so you simply will not see a 20MHz signal on it.

If you can get a watch crystal of 32,678Hz and try with higher resistance values to match that sort of crystal (suggested above) then you might see it.

pc1966 said:

Looking up your DSO138 oscilloscope it appears to have a 200 kHz bandwidth so you simply will not see a 20MHz signal on it.

If you can get a watch crystal of 32,678Hz and try with higher resistance values to match that sort of crystal (suggested above) then you might see it.

Click to expand...

Very good idea and I didnt check it myself or try to learn this little detail about my scope. Very good that you check it for me. Excellent. Yes this make sense. The scope may be very much, very limited. As I anticipated without actually check it like you did. Like I said, very dinky. Haha. Hmmm, that makes a ton of sense. I will TRY to find - if im lucky something under 200kHz in my OSX (OScilator X=value) known values bag.

I have an Ingeeneerious idea !
Is there a way to SPLIT the fv of a BIGger OSX into a smaller fv? Just to bring it down under 200 kHz? Just enough to measure it with my dinky osc? Ha?