This is NOT a Peltz sinusoidal oscillator, but...

[Sacodepatatas]

Hi there,

I've been checking this thread,

https://www.eevblog.com/forum/projects/simple-sinusoidal-oscillators/
and began to play with the Peltz oscillator. But at certain moment I realized that a Peltz has a similar topology than a Schmitt trigger made of a differential pair that feeds back positively, and the LC tank tunes the feedback ratio up to the resonance frequency.

So, if the regular Schmitt trigger and the Peltz are siblings, what about THE OTHER Schmitt trigger topology? The one using a complementary BJT pair, that is often used for modeling an SCR? Yes, I derived an oscillator from this last one, and I'm really surprised of its performance. The LC tank is placed between both collectors, and this setup has pros and cons. Maybe a picture worths more than 1000 words...



You can see that there is distortion at the collectors of seemingly both outputs, BUT, one cancels out the other's distortion, and the differential signal is very clean. Why? Let's consider the base currents negligible. What we know:

-Any BJT working in active zone has a very high collector impedance. The collector current is nearly inmune to any external influence other than the base current.
-If a BJT saturates, the gain decreases and its collector has no longer high impedance.

When the oscillation builds up, one of the transistors begins to weak saturate, but the other transistor is well in active region. This means that its collector will track the voltage of the saturating transistor (because its collector current cannot change externally), and the difference voltage is the natural resonance of the LC tank. The higher the Q factor of the tank, the more pure the sine wave is being generated, and it can be really good.

The only drawback that I see is that its output is differential. But I think it's a cheap price to pay for what we get.

I guess that nowadays every simple circuit has been reinvented a thousand times, but in case it has not yet a name, I'll call it the "Patata Oscillator" or "Patatator" (I think this last one rocks!).

I also incluye pictures of my breadboard testing...

[RoGeorge]

Nice potato oscillator you did there! 

[Sacodepatatas]

Nice potato oscillator you did there! 

Thank you. I did it just as a proof of concept, but I didn't expect this performance. To be honest, at first I was dissapointed because I didn't consider the ESR of the coil and capacitors, hence the circuit was working on simulation but not on breadboard. Then I set realistic values of ESR and that let me tune the simulation, then tried those resistor values, and voilà, that pretty sinusoid appeared on the oscope screen. I love that symmetry not just aesthetic, but also functional (as explained). I also tried low Q  LCs, and while the single ended outputs were heavy deformed, the differential was still a recognizable sinusoid (although not perfect)

Pros:
-High output due to the differential voltage.
-Non linearities dissapear by design. Main contribution to the THD comes from the LC tank itself.
-There is an compromise between quality of the sine and voltage output. For the same L, increasing the C Will not only decrease the frequency, but also the amplitude. But you get beautiful sines at each collector. In single ended it behaves like the Peltz. The circuit also oscillates badly and wild without the capacitor.

Cons:
-Needs a balun or a differential amplifier in order to get the most of this setup.
-Increasing VCC decreases Vout. The same circuit, at 5V gives less than half the Vp than at 3.3V. But i think it's because of the transconductance of the transistors. With ones of higher transconductance, you should get almost rail to rail output even at 5V.

I'll experiment further with this "potato oscillator".

Cheers.

[moffy]

Nice variation on the Peltz, I like the 'Patata' name, nice.
Film capacitors have a low dissipation therefore high Q but inductors can be problematic with their series resistance dampening the Q at resonance.

[doktor pyta]

Wind another winding on L1 to easily get output signal from and who knows- maybe it can be useful.

[mawyatt]

Interesting adaptation on the Peltz, nice

Best

[Sacodepatatas]

Wind another winding on L1 to easily get output signal from and who knows- maybe it can be useful.

Yes, that's a balun, or balanced to unbalanced coupling transformer. It can be a simple solution but coupling transformers also have their problems. If the oscillation frequency is low enough (below 100kHz), I prefer a differential amplifier. If RF, of course a coupling transformer is preferred.

[gbaddeley]

R3 or R4 is redundant, as they are both in series with the power supply?
What happens if the R approaches zero ohms? Or a very high value?

[Sacodepatatas]

R3 or R4 is redundant, as they are both in series with the power supply?

At first sight, you'd be right, but let's see why are these resistors placed in there...

A resistor placed at the emitter of a BJT sets an amout of negative feedback. The current through the R shifts the emitter voltage so the base will be less polarized. In certain situations we would like to control how much total feedback we get by setting a little amount of negative feedback.

In Spain we have the aphorism "Las gallinas que entran por las que salen", that means "the chickens that enter, replace the ones that exit". Electrons flow from GND to Vcc, they don't have any other route to follow. However, the LC tank is a temporary Energy storage. That means that the electrons may not flow always at the same rate. Voltage at the collectors can be driven crazy. If that happens (when the oscillator amplitude surpasses the supply margins), both transistors can run out of active zone at the same time, and then the LC tank will no longer see everytime a high impedance. The wave gets distorted because part of the energy in the tank gets leaked.

Resistors at both emitters ensure that both transistors will be kept in active zone most of the time. In fact, a small amount of variation of one of the emitter resistors will ensure which transistor will mildly saturate. You don't usually want a hard saturation.
The funny thing is that if you remove one of the resistors and double the other's value, then you ensure that the transistor without the degeneration resistor will saturate hard, and both outputs get hardly distorted. But, because the other transistor is still in active mode, the differential voltage still holds and the output would still be a nice sine wave. Also, keeping output voltages symmetric from Vcc/2, and within supply range, is something desirable if you want to feed them into an opamp, for example.

What happens if the R approaches zero ohms?

Then there's is no significant amount of negative feedback, the both transistors will hard saturate, the wave will reach Vpp->2*Vcc, and you have an awful output.

Or a very high value?

If your LC tank had 0 ohm of total ESR (that is, an infinite Q factor), then the value of R at the emitters would not be so critical. You could set, R1=R2=100k, and R3=R4=10k. If you try these values in a simulator, the oscillator will still work:

https://www.falstad.com/s.php?s=RMFLVd

But with realistic values of ESR (my inductor should have ESR around 60-70 mOhms, and that capacitor about 10-20). With these values, R3+R4 above 220 will kill the gain. The emitter resistor values are a compromise between Vpp output, low distortion, stability under ambiental variations, and linearity.

[gbaddeley]

Regarding R3 & R4. I was thinking that one of them would be redundant if  the output was taken from a separate winding on the inductor?
A concern would be that the parasitics to ground would then be unsymmetrical and increase distortion or reduce stability?

[Sacodepatatas]

Regarding R3 & R4. I was thinking that one of them would be redundant if  the output was taken from a separate winding on the inductor?
A concern would be that the parasitics to ground would then be unsymmetrical and increase distortion or reduce stability?

I think the asymmetry would be interesting if you wind a separate winding, not for output, but for RF input. Then, Re would be a trim pot for controlling the oscillation threshold of a regenerative detector, and the detected output could be taken directly from the emitter.

[RoGeorge]

R3 or R4 is redundant, as they are both in series with the power supply?
What happens if the R approaches zero ohms? Or a very high value?

Interesting question(s), this made me think of a minimalistic version, one with only one of each:
- one battery
- one PNP
- one NPN
- one L
- one C
- one R
with one unbalanced output, that gives one point one Vpp at one kHz and less than point one % THD. 


(it's the same schematic but rearranged and with some resistors removed, so a minimalistic potato osc running as a low THD audio generator from a single cell NiMH battery  )





This is one of a kind performance, so It just became one of my favorite oscillators, and I think one day I'll build one.  Take a shot every time you see one! 

[Sacodepatatas]

Very clever minimalistic version by playing with Vcc to control the base currents and saturation of the upper transistor 😃. I wonder what part of the contribution of the THD belongs to the LC tank, and what of the non ideal.behaviour of the transistor in active zone. If the later, using transistors with high HFE and V(early) should improve the THD figure.

Definitively i want to experiment further with this oscillator 😁.

[Zoli]

...
This is one of a kind performance, so It just became one of my favorite oscillators, and I think one day I'll build one.  Take a shot every time you see one! 

I foresee a lot of "fun", especially to find the proper inductor; a close match for the simulated inductor: https://www.hammfg.com/part/195P5?referer=949 ; 5.9 lbs (2.68 kg)!!!
IMO, this oscillator require relatively high Q inductors to work; low(=audio) frequency with high Q means big size. Otherwise, it's a nice idea, with (my)estimated optimal frequency range between 0.1-10MHz.

[MariuszD]

This circuit is an oscillator with negative resistance. Besides the transient simulation, it is worth doing a simulation of how the current depends on the voltage in place of the LC circuit.

The negative resistance can and should be adjusted to the resonant circuit.

Compared to the Peltz oscillator, in this circuit, the negative resistance is constant over a wide range of voltages, but the amplitude of the oscillations is similarly limited by the conduction of the BC junction.

[RoGeorge]

I foresee a lot of "fun", especially to find the proper inductor; a close match for the simulated inductor: https://www.hammfg.com/part/195P5?referer=949 ; 5.9 lbs (2.68 kg)!!!
IMO, this oscillator require relatively high Q inductors to work; low(=audio) frequency with high Q means big size. Otherwise, it's a nice idea, with (my)estimated optimal frequency range between 0.1-10MHz.

Mine is about a thousand times lighter than 2.68kg. 



I've used a toroidal 2x5mH Common Mode Power Line Choke (similar with this one here:  https://www.we-online.com/components/products/datasheet/744821150.pdf ).  They have two identical sections of 5mH each, both sections on the same toroid.  These are used as mains filters, very common in PC power supplies, TV power supplies, etc.  Mine is from a broken PC power supply.

Because the inductance of a coil increases with the square of the number of turns, when the two sections are put them in series they measure about 28mH (or at least that is how much my T4-LCR tester is showing).  That is why I've used the 28mH value in the simulation.

Same with the 680nF value, it happened to have a big (polypropylene?) 0.68uF/100Vdc (fabricated Dec 1979  ).  For minimum distortions, a rolled capacitor is supposed to work better than a much smaller ceramic capacitor.  Together with the yellow choke in the pic, they oscillate at about 1kHz.

[RoGeorge]

I wonder what part of the contribution of the THD belongs to the LC tank, and what of the non ideal.behaviour of the transistor in active zone. If the later, using transistors with high HFE and V(early) should improve the THD figure.

This is the THD measured for a BC547B + BC557B + 28mH + 0.68uF + 600ohms + 1.2V NiMH.


THD -42.9dB (0.7%) at the "normal" (for this type of oscillator) audio level, -6dB, and about 1.1kHz


THD -60.0dB (0.1%) when the osc amplitude is lowered down to about -20dB audio level

The number in lower-right corner is the THD, just that it is in dB instead of %, to convert it:

Code: [Select]

$ qalc
> 10^(-42.9795/20) to %
  10^((−42.9795) / 20) ≈ 0.7096186157%

> 10^(-60.0119/20) to %
  10^((−60.0119) / 20) ≈ 0.09986308999%

First one is for a "normal" level, which is about 2x0.6Vpp so about -6dB audio, and the measured THD is 0.7%
The second screen capture is for a smaller signal.  I've increased the resistor in the emitter, to reduce the amplitude of the oscillations until they become af about -20dB.  At a level of -20dB, it gives about 0.1% THD.

The load was the desktop's sound card, both channels of the line-in from PC in parallel with the LC tank.  The oscillations amplitude is heavily dependent of the load, I had to use a trim-pot to adjust the osc.  Needed value are nowhere near 100k (as it was in the LTspice simulation).  In practice it needs something like 1..10k, so to give a decent Ic for transistors.

It was a surprise to find out that in practice the amplitude can stay constant even at levels much smaller than the expected sx0.6Vpp.  The amplitude can stay constant even at -20dB (instead of "normal" -6dB), by increasing the resistor trim-pot in the emitter.


About distortions, the base-collector junctions of the transistors are in parallel on the LC tank, each BC junction is limiting half of the sinusoid.  Heaving the same forward Vbc for both transistors would help minimizing the second order harmonics.  Similar, heaving the same beta for both transistors would help.

If the resistor is too small, then the amplitude of the oscillation will tend to rise, but the BC junctions will cap it at 0.6V, making the waveform to look like more and more like a square wave, which will drastically increase the odd order harmonics.

[Sacodepatatas]

If the resistor is too small, then the amplitude of the oscillation will tend to rise, but the BC junctions will cap it at 0.6V, making the waveform to look like more and more like a square wave, which will drastically increase the odd order harmonics.

Yes, the BC diode starts conducting when the transistor begins saturating and that limits the wave to increase forever. The idea is that the emitter resistor will give us enough margin to cross the boundary where gain is slightly above 1 (working in active mode) to slighly below 1 (the first transistor begins saturating). If the same setup was done with mosfets, i think the body diode would cap the wave amplitude 0.7 volts ahead supply rails.

[mawyatt]

If the resistor is too small, then the amplitude of the oscillation will tend to rise, but the BC junctions will cap it at 0.6V, making the waveform to look like more and more like a square wave, which will drastically increase the odd order harmonics.

Yes, the BC diode starts conducting when the transistor begins saturating and that limits the wave to increase forever. The idea is that the emitter resistor will give us enough margin to cross the boundary where gain is slightly above 1 (working in active mode) to slighly below 1 (the first transistor begins saturating). If the same setup was done with mosfets, i think the body diode would cap the wave amplitude 0.7 volts ahead supply rails.

One of the interesting design methodologies we utilized decades ago for low noise oscillator design was looking at the integral of the incremental gain in a complete single waveform cycle. For Steady State this must be exactly unity, otherwise the oscillator isn't in SS. All oscillators must be non-linear and have a gain >1 at startup for the waveform to buildup, then reach SS when the gain decreases to exactly unity.

An interesting side note from way back then was old school thinking was in low noise power amplifier design the amp must be highly linear, some research was going on with highly limiting amp design where the output transitioned from full cut-off to saturation as quickly as possible (think square waveform). The idea was the active device contributed almost no noise while cut-off and in deep saturation, and only contributed noise during the transition "linear region" between cut-off and saturation and visa-versa. Vaguely recall this thinking was also going on in some low noise power oscillator designs.

Best

[RoGeorge]

I think I've seen something like that in a paper about some nW, or maybe it was pW, don't recall, but very, very low power, much lower than "normal" oscillators.

That particular design was for 32768Hz resonators, the kind used in wrist watches or in microcontrollers that includes an RTC.  The idea was that they were applying a very short pulse, then they were disconnecting the resonator from the amplifier, and let it run freely for many cycles.  Because such resonators have very, very high Q, the ringing was lasting a lot from a single pulse, and when the amplitude was decreasing too much, they were pulsing it again with a spike.

Since the entire circuit was aware of the resonator's phase, it was possible to place the spike at just the right moment, such that the new drop of energy will come in phase with the already existing oscillation.  Same with the quantity (amplitude) of the spike, they knew how much the resonator was decaying since the last spike, so they were pouring just the right amount of energy into the resonator.

The overall effect, aside from being very low power, was a much better phase noise, and better frequency accuracy.  It was all very cleverly implemented, with only a few CMOS analog switches controlled by a couple of CMOS gates and flip-flops.  The amplitude comparator was also implemented with logic gates.

[mawyatt]

Very clever implementation indeed!!

Best

[RoGeorge]

There were a few variations around those ideas, this is one of the implementations:



If anybody curious about what exactly was that about and what else was there, found the paper:
Ultra-Low Power 32kHz Crystal Oscillators: Fundamentals and Design Techniques - DOI 10.1109_OJSSCS.2021.3113889

[PCB.Wiz]

This was the low-power oscillator idea that impressed me the most

https://www.radiolocman.com/shem/schematics.html?di=105602
(link fixed)

It uses PNP/NPN pair, but includes a clever dc restoration-agc feature in the BE junctions, so the current reduces once it starts.
That makes it very Vcc tolerant, and also means the active devices have very short ON conduction on each peak, so everything remains balanced.

[Sacodepatatas]

This was the low-power oscillator idea that impressed me the most

https://www.radiolocman.com/shem/schematics.html

Is the link right? What I see is just a listing of projects.

Browsing that index I found several wien oscillators, each implementing its own way of stabilizing the oscillation. But i don't know if it's me or what, but I feel that everybody's implementation is failing the obvious way. Let me explain:

A simple wien oscillator is a passive bandpass filter that attenuates with a peak gain of 1/3 at center frequency. The obvious thing is that the frequencies far from the center are attenuated much more than 1/3. Then why every circuit is taking the oscillator's output from the filter's input instead of the filter's output, which always, in more or less amount, will be cleaner than the input? What i'm missing?

I attach an example.

Edit: Ok, most of  the gain limiters are after the opamp's output, so then the opamp is amplifying the clean output of the filter. Forget the above 😅

[PCB.Wiz]

Is the link right? What I see is just a listing of projects.

oops, fixed above.