LTSpice Model of LA3A Autoformer

I want to do an LA3A circuit (no I/O, just the attenuator resistor network with T4B cell, EL panel driver w/ autoformer and associated controls).

The B11184 autoformer looks like a convential 5W 70V line distribution transformer:



so I was going to use this from Parts Express:

http://www.quamspeakers.com/filebin/product_spec_docs/spec-pdfs/TBLU.pdf

But just to be sure I understand what this part is doing I thought I would model things in LTSpice:



I have no idea what the real inductance values are for each tap would be for a line transformer like this but I found that using values relative to the wattage of each tap (but in reverse order) yields the right power into an 8 ohm load (meaning if I put the stimulus on the 5W tap and measure 5W power on the load and then switch to the 0.31W tap, it now gives 0.31W through the load, so it's correct).

So now I want to see what the step up is. For some reason I don't get the same results of I connect to the 0.31W tap vs using a separate winding equal to the sum of all of the windings (5+2.5+1.25+0.63+0.31 = 9.69H). If I use the tap I get a step-up of~7.25x. If I use the separate 9.69H winding I only get square root of 9.69/0.63 = 3.9x.

So using the 0.31W tap that steps up 12V to ~47V. I use the separate 9.69H winding that steps up 12V to ~87V.

So how does one model a 70V line transformer like this in spice?

Does anyone know what the step up is supposed to be in an LA3A?

squarewave said:

I want to do an LA3A circuit (no I/O, just the attenuator resistor network with T4B cell, EL panel driver w/ autoformer and associated controls).

The B11184 autoformer looks like a convential 5W 70V line distribution transformer:

Click to expand...

But probably much higher nominal inductance.

I have no idea what the real inductance values are for each tap would be for a line transformer like this but I found that using values relative to the wattage of each tap (but in reverse order) yields the right power into an 8 ohm load (meaning if I put the stimulus on the 5W tap and measure 5W power on the load and then switch to the 0.31W tap, it now gives 0.31W through the load, so it's correct).

Click to expand...

This is correct, except the COM should be connected to GND. It looks like the inductance values are adequate (I mean by that that in reality, they probably are within -50/+100%).

So now I want to see what the step up is. For some reason I don't get the same results of I connect to the 0.31W tap vs using a separate winding equal to the sum of all of the windings (5+2.5+1.25+0.63+0.31 = 9.69H).

Click to expand...

That's because windings in series on the same core do not add algebraically, they add quadratically.
1H+1H=>4H

So how does one model a 70V line transformer like this in spice?

Click to expand...

More or less like you've done, but keeping the quadratic addition in mind.
BTW LTspice does not like K=1. Use 0.9995

I have succesfully tried an Edcor XSM 1:4 and an ordinary small Hammond power transformer (6 VA, roughly 1:5) in my Franken-2A/3A build with a homebrew T4B cell. None of those is an autoformer but connecting the primary in series with the secondary is basically the same. Both worked fine but as I needed the Edcor elsewhere, the Hammond made it into the finished unit.

You don't need a lot of power to drive the EL panel as it's mainly a capacitive load. This load and the inherent frequency response of the panel itself also mess with the frequency response of the actual compression. The EL panel mostly reacts to frequencies around 200~800 Hz - much, much less to bass and treble. As long as the transformer has a halfway decent bandwidth, you're good to go.

For simulation I used this setup in LTSpice. The values may be way off from the actual components but it worked surpisingly well against the real unit.

Here's my model so far:



Although I am going to assume that it's not entirely correct at this point because I get some strange results.

For example, the 4.7n HF contour cap only boosts frequencies well above 1 KHz (+3dB starts at 3KHz). Maybe it only kicks in when the circuit clips and the higher frequencies are actually affecting the duty cycle? Or maybe I need to model the panel?

I don't have spice models for 2N3053 / 2N4037 either which I have to guess are pretty slow, low gain and probably have some current limiting behavior to them. That could matter.

What is the pre-regulated voltage? Is it really 40V? This schem:

https://www.dropbox.com/s/6ofhb13fweofqnk/SA-3A_REV2_SCHEMATIC.pdf?dl=0

says 35V (although it also claims post-regulated voltage is 28V which seems high to me). In my model clipping is asymmetric below 40V. Either the voltage is wrong or the bias is wrong.

The grounded emitter of Q10 seems a little reckless.

I wonder if the engineers really thought about this one or if they just tinkered until it gave them the desirable behavior.

squarewave said:

For example, the 4.7n HF contour cap only boosts frequencies well above 1 KHz (+3dB starts at 3KHz).

Click to expand...

I don't see anything wrong with that. It's probably meant to compensate the HF loss resulting fron the transformer (inductive) driving the load (capacitive).

What is the pre-regulated voltage? Is it really 40V?

Click to expand...

With the 30Vac xfmr, it seems correct; in fact it is probably a little higher because the xfmr is lightly loaded.

This schem:...says 35V (although it also claims post-regulated voltage is 28V which seems high to me).

Click to expand...

No; 35V is one of the regulated voltages.

In my model clipping is asymmetric below 40V. Either the voltage is wrong or the bias is wrong.

Click to expand...

Operating point in this stage is dependant on Q10's gain. Resistors R35 & 33 must be adjusted for symetric clipping.

The grounded emitter of Q10 seems a little reckless.

Click to expand...

Why? I don't see anything wrong there.

I wonder if the engineers really thought about this one or if they just tinkered until it gave them the desirable behavior.

Click to expand...

That's a strange comment; it's the usual procedure: think about it, build it, and then, "tinker" till it does what you want.

squarewave said:

Or maybe I need to model the panel?

Click to expand...

Not just this!

Remember, this compressor is based on a feedback topology - the signal in the sidechain highly depends on the output signal of the compressor, both level and frequency wise. The sidechain circuit reacts completely different, when it's actually compressing something. Trying to model this in Spice is getting very complex very quick.

When I built my unit, I had a working circuit in front of me on breadboard. I used LTSpice for the different circuit sections (modeled separately) helping me understand what everything does. This way I could try component changes before applying it to the actual circuit. Be aware that the voltage on the autoformer's secondary is getting very high! It's no fun accidentally touching it when it passes audio... 

One of the things I was able to improve about the original LA3A sidechain was converting the output section to Class AB operation as opposed to Class B in the original. It just needed two diodes and a different resistor value. The distortion induced by Class B might not do any harm to the sidechain signal but it finds its way into the ground system and with that into the actual audio signal - Class AB operation fixed that... 

fripholm said:

The distortion induced by Class B might not do any harm to the sidechain signal but it finds its way into the ground system and with that into the actual audio signal - Class AB operation fixed that... 

Click to expand...

Mmm, that is interesting. I wonder if simply isolating that ground and having a separate ground wire back to the first filter cap (C10) might also fix that.

abbey road d enfer said:

This is correct, except the COM should be connected to GND.

Click to expand...

How do you know for sure COM is connected to ground?



The schematic pretty clearly says '5'.

Using the 5W tap yields an 18% increase in output voltage because it's only using one segment of winding (between 2.5 and 5) whereas using COM includes two segments.