Output stage oscillating

I have rather interesting issue with the output stage of one of the devices I'm working on right now. I've atached the schematic of the output stage below. Output transformer is CineMag CMOQ-2

Here's the problem. When I connect the transformer in 1:1 configuration, everyting is fine. However, when I connect the transformer in 1:2 step-up configuration, I get a huge oscillation on the output at 300 kHz, several volts amplitude. I even tried 1:3 configuration and got the same oscillation, but at 150 kHz. 600R resistor on the secondaries doesn't help.

Why would that be? What can be done to cure that? I really need a transformer t work as 1:2 step-up here.

It is almost certainly because the primary of the transformer is in one of the  the feedback loops. This circuit looks a bit like a transistor version the negative impedance circuits commonly used with low cost transformers that have a low primary inductance. The Cinemag, being somewhat better, probably has a higher primary inductance than was intended for this circuit. What transformer is specified for this circuit?

Cheers

Ian

Ian,
The actual transformer specs are unknown. It was probably wound in-house by the manufacturer. I'm guessing the 1:2 ratio from other indirect clues.

Stability criteria for negative feedback circuits involve the NF being attenuated below unity, by the frequency when forward path delays equal 180' of phase shift causing NF to look like positive feedback.

In addition, loading of the output stage working with the open loop source impedance of the stage can cause phase lag that hurts stability.

The evidence that 1:2 behaves worse than 1:1 could be interpreted two ways. 1:2 voltage ratio is a 1:4 impedance transform so loading on the output stage is 4x worse. A second consideration is that any leakage or crosstalk from the transformer secondary is now 2X the voltage.

I'd be suspicious of the positiver feedback coming back into the emitter which is increased by the 1:2 connection. (as others have noted).

JR

Aside from fog on NFB/PFB and low/high output Z, it is class A with not much peak current ability.

I would seriously add a chip buffer. '5532 can be run on zero and -16V, unity gain, drive less than 300 Ohms without shame. At only 16V supply one section can almost drive 150 Ohms to 6V peak, which is about as much as this circuit can do in 600 Ohms.

that Cinemag xfmr is an API OPT, quad-filar wound, low inductance, (depending on core, they ship 80Ni ,50/50 and steel, steel works best)

there is some clever design work here, that quad wind has a ton of capacitance from wind to wind, the pri and sec wires are right next to each other, as a result, you get frequency response that extends way past the audio range,

the penalty for this is increased phase shift at the high end, at 1 M Hz it is about 180 degrees out, so we suspect that your problem relates to the nature of the design, when you series the secondaries, you are probably altering the phase shift due to changes in capacitance,

edit: phase shift will be on the secondary, which is not in the feedback circuit so shift would not seem to be a problem, what were we thinking?

are you grounding one of the secondary leads or running it balanced? try both ways.

interesting schematic, they are hooking the B+ to the shield, never seen this before, there is  no shield on the Cinemag so you can not try this.

you might try using just one winding for the primary,  and two in series for the secondary, or if that is what you are doing, use two primaries in parallel and see if it helps,

spec sheet for the cinemag-notice the grounding scheme on the cmmr test setup >

interesting phase curves -  Ni vs Steel,

at 300 K Hz you have about 60 degrees, is that a critical value? might be, have to read RDH4,

why the difference in amplitude with different cores? probably different inductance combining with fixed capacitance,

always wondered why steel sounded a tiny bit different (better?) than Ni (we have dual API's in a rack, one nickel, one steel) ,
now we know, different hi end curves,

what's with the jagged lines with Rs=50?

are those the lams talkin?  Ni looks cleaner than steel,  magnetic domains more evenly distributed in Ni?
or is it just smoke coming out of  ;D

Ilya said:

I have rather interesting issue with the output stage of one of the devices I'm working on right now. I've atached the schematic of the output stage below. Output transformer is CineMag CMOQ-2

Here's the problem. When I connect the transformer in 1:1 configuration, everyting is fine. However, when I connect the transformer in 1:2 step-up configuration, I get a huge oscillation on the output at 300 kHz, several volts amplitude. I even tried 1:3 configuration and got the same oscillation, but at 150 kHz. 600R resistor on the secondaries doesn't help.

Why would that be? What can be done to cure that? I really need a transformer t work as 1:2 step-up here.

Click to expand...

It's a negative-impedance amplifier, based on positive feedback. The ideal amount of positive feedback is when the negative impedance of the circuit cancels the positive impedance of the load.
It may very well be that when you connect the primaries in parallels, you go too far in impedance compensation, tehn the circuit becomes an oscillator.
In fact, this circuit works rather well for compensating the purely resistive term in the load impedance, but, as you know, the load is complex, in particular the parasitic capacitance (the one that reflects as a cap in parallels with the primary) makes the positive feedback increase with frequency.
In order to compensate that, you may try reducing the HF loop gain, by shunting the NFB resistor R28 with a small cap until the oscillation disappears. Other possibilities are a small cap between c and b of TR8, a less small cap across R31, or an inductor in series with the primary.
Since the most important data concerning with this issue are missing (stray capacitance, leakage inductance) it's impossible to predict/quantize fixes.

Thanks gents for your input!

I'm going to try out what abbey road suggested since this is the simplest thing to do. I don't want to add a 5532 buffer as PRR suggested, I'll leave it as a last resort measure. This beast is on the PCB already, and the space is rather tight even for a single chip.

I've also thought about the base stop resistor at TR8, something around 100-1000R. Do you think it's worth it?

Edit: CJ, I'm not grounding any windings. And grounding secondaries didn't help when I tried it. It's curious to see a ~300kHz spike in the datasheet. Looks like this thing is at work here. I also tried single primaries and 3 secondaries in series - the osc is still there, but at 150 kHz IIRC.

I am not convinced the Cinemag needs the positive feedback. I would be tempted to simply disconnect pin 3 of the transformer and connect it direct to 0V .

Cheers

Ian

ruffrecords said:

I would be tempted to simply disconnect pin 3 of the transformer and connect it direct to 0 .

Click to expand...

Simpler to just short out R33.

togology is similar to several of the RCA germanium era including the BA-31c,
sn 50xx fitted with UTC CR-725 (100:600).
return of transformer primary winding grounded.  germanium issues aside, amps are reasonably stable.

gridcurrent said:

togology is similar to several of the RCA germanium era including the BA-31a.
return of transformer primary winding grounded.  germanium issues aside, amps are reasonably stable.

Click to expand...

Yes, the output stage is just an emitter follower with a constant current load. Rough in the head calc indicates the output stage quiescent current is around 12mA. Peak to peak output current should be up to 24mA which is about 8mA rms which into a 200 ohm load is just over 10mW or about +10dBm.

Cheers

Ian

Ok. Quick and dirty test showed that both solutions work. Inserting 10p cap parallel to R28 stops the osc. As well as shorting the 8R2 resistor does.

The question is which solution is better?

Ilya said:

Ok. Quick and dirty test showed that both solutions work. Inserting 10p cap parallel to R28 stops the osc. As well as shorting the 8R2 resistor does.

The question is which solution is better?

Click to expand...

The 10pF cap introduces a one-pole LPF well above the audio passband, so not audible.

Shorting 8R2 defeats the circuit operation.

IMO between those two options adding the 10pF is preferred.


JR

JohnRoberts said:

Shorting 8R2 defeats the circuit operation.

Click to expand...

Circuit 'operation' depends on the transformer but also what you connect to the other side

I first saw this at Calrec where it was used to improved the already good performance of a big Lundahl.  Ken Farrar might have copied it from Lundahl as their patent came out at about the same time.

But there was precise matching between the circuit values and the transformer.

ricardo said:

Circuit 'operation' depends on the transformer but also what you connect to the other side

I first saw this at Calrec where it was used to improved the already good performance of a big Lundahl.  Ken Farrar might have copied it from Lundahl as their patent came out at about the same time.

But there was precise matching between the circuit values and the transformer.

Click to expand...

Yes, the evidence that it worked for one winding connection and not the other suggested that the values need to be different for the lower impedance connection that oscillated. Changing the 8R2 resistor value without bench testing to dial it in may not be straightforward for simple DIY.

JR

gridcurrent said:

togology is similar to several of the RCA germanium era including the BA-31c,

Click to expand...

I wouldn't say that; there are two major differences. The negative-impedance drive is one. The other is the positive feedback from the collector of the upper transistor to the base of the lower; this stage does not operate on constant-current, it is in fact a real push-pull, in a way similar to the SRPP topology used with vacuum tubes. The resulting output drive capability is much higher than the basic assessment of idle current would suggest.
That approach was quite common in the early days of SS electronics.

JohnRoberts said:

Yes, the evidence that it worked for one winding connection and not the other suggested that the values need to be different for the lower impedance connection that oscillated. Changing the 8R2 resistor value without bench testing to dial it in may not be straightforward for simple DIY.

JR

Click to expand...

I've got a good scope and signal generator on my bench. Would that be enough? Do you have any suggestions regarding the dialing in the correct value of that resistor?

Ilya said:

I've got a good scope and signal generator on my bench. Would that be enough? Do you have any suggestions regarding the dialing in the correct value of that resistor?

Click to expand...

No, that wouldn't be enough. The "ideal" value is the one that gives the lowest LF distortion, but one has to account for variations due to temperature. So you may find the point of lowest distortion, but find it changes with temperature, so you need to backup a little.
The oscilloscope is good to detect the onset of oscillating, though.
That type of topology is brought to an epitome in the Audio Precision oscillator, where the positive feedback loop includes a temperature dependant resistor, in order to compensate for the different tempco of copper (xfmr windings) and metal oxide (resistors).