dual differential input stage

[bcarso]

Saw this just now:

http://www.edn.com/blog/1700000170/post/400007840.html?nid=2433&rid=483918643

One wonders whether there would be any point in extending the concept to higher multiples.

 

[Guest]

[quote author="bcarso"]Saw this just now:

http://www.edn.com/blog/1700000170/post/400007840.html?nid=2433&rid=483918643

One wonders whether there would be any point in extending the concept to higher multiples.[/quote]

My turn: twin complementary diffcascade.



My turn #2: differential cascade using Tango pairs (FET+ BJT with resistor in emitter)

My turn #3: complementary #2

 

[Guest]

My turn #4: a single transistor (feedback to emitter) == no crossover point at all. However, a DC shift between inverting and noninverting inputs...

My turn #5: complementary #4. Feedback through equal resistors to both emitters, == no DC shift.

 

[JohnRoberts]

Interesting. Looks like potentially lower noise than resistive degeneration. Curious to know how it works in practice with NF loop closed. Paralleled input stages will have more transconductance so may need larger comp cap for stability. How that plays out may indicate how much benefit to HF linearity this might offer, but it seems more linear input range should be a good thing.

If this has utility it could be exploited nicely in integrated opamps, but I'm not smart enough to speculate. Many oddball input differentials seem like they're trying to reduce transconductance.

Thanks for the link.

JR

 

[clintrubber]

[quote author="JohnRoberts"]If this has utility it could be exploited nicely in integrated opamps, but I'm not smart enough to speculate. [/quote]
Has already been used, for instance for extending the input-range
of transconductors in fully integrated gm-C filters.
Indeed very useful.

Regards,

Peter

 

[JohnRoberts]

[quote author="clintrubber"]
Has already been used, for instance for extending the input-range of the balanced transconductors in fully integrated gm-C filters.

Regards,

Peter[/quote]

I'm not familiar with those, but good point. extending the linear input range has merit for simple OTAs like classic 3080 which were routinely used open loop, so stability compensation is not an issue.

JR

 

[recnsci]

[quote author="bcarso"]
One wonders whether there would be any point in extending the concept to higher multiples.[/quote]

I dont see why not. Preferably, you should use some more advanced tool
for higher multiples. Like Matlab. You could write nice optimisation code that
would do all the hard work, instead of hit-and-miss Excel aproach.

cheerz
urosh

 

[recnsci]

[quote author="Wavebourn"]
My turn #2: differential cascade using Tango pairs (FET+ BJT with resistor in emitter)
[/quote]

What's Tango pair?

cheerz
ypow

 

[PRR]

> Saw this just now:

The transfer curve is remarkably like a Fender 5F6A output stage's. Underbiased (cool) heavily loaded push-pull 6L6 with a zero kink and end-bend.

> One wonders whether there would be any point in extending the concept to higher multiples.

Extension of "linear" range. Strict tradeoff between bump size and number of devices. Tschebyshev filter bump analysis might apply, if you bent Tscheby to another angle and had a big hammer. Or SPICE will give answers (maybe useful answers) to such questions in a jiff.

> Looks like potentially lower noise than resistive degeneration. Curious to know how it works in practice with NF loop closed. Paralleled input stages will have more transconductance

If design is constrained by input bias current (common in simple BJT work), noise and GBW are unchanged.

If you have a single pair of Hfe=100 devices and a 10uA input current spec, you run them at 1mA each.

For the same devices and input current, the 2-pair form would be biased at:

0.66mA 0.33mA
0.33mA 0.66mA
===1mA ===1mA

Noise and Gm unchanged, die area doubled, linear range nearly doubled. (GBW may fall due to more collector capacitance at the upper mirror; or not.)

In a 10-pair stack, at any given point in time or input voltage, 8 of the 10 pairs are inactive. One side cut-off, the other side taking full tail current but giving zero Gm because tail is infinite impedance (near enuff).

It is interesting that it does seem to "switch" smoothly for input ranges far larger than we expect from BJT. With ripple, but I suspect that closer offsetting can meet any finite gain-smoothness spec (at cost of more devices). May not be so easy to fool the ear.

For common processing there is a 7V limit for breakdown of the very-off junctions. But that needs like 150 pair! But 300 devices is nothing in this day of million device CPUs. But good audio devices are not small. But if you can budget 300 devices, you can budget a few diodes. (And if you get off commodity Silicon: many older devices and Ge devices had huge emitter breakdown voltages.)

EDIT No, goldammit, ALL the full-on devices suck full base current. Total input current rises with number of pairs. Good stuff like Gm does not improve. So you'd need an unusual application to accept the lousy DC spec. I suppose you could buffer the input, but that's noise and another pole....

 

[clintrubber]

[quote author="PRR"]EDIT No, goldammit, ALL the full-on devices suck full base current. Total input current rises with number of pairs. Good stuff like Gm does not improve. So you'd need an unusual application to accept the lousy DC spec. I suppose you could buffer the input, but that's noise and another pole....[/quote]
The 'trick' can be done for FET-diffpairs a well, avoiding that 'base' current penalty.

Regards,

Peter

 

[PRR]

> The 'trick' can be done for FET-diffpairs as well

Or thermionic valves, which are just hot sloppy FETs.....

Just because Shockley's 26mV is more like 1V in a tube is no reason not to want more input range. (But doubling-up bottles gets expensive much faster than doubling-up sand chips.)

> avoiding that 'base' current penalty.

There's always a grid current.

Tubes and FETs, the DC current may be too small to hurt us.

But the input capacitance is often annoying. Even in guitar amps, you can't see a grid as 1 Meg, you have to see ~~100pFd, which dips toward 100K impedance in "the audio band", and limits guitar amp plate resistors and mix/tone networks to ~250K. Hi-Fi designs have to work at lower impedances, lower gain per stage.

Small FETs can be quite thrifty. But Cin, Gm, and Imax all go together. And in many sand-state FETs, Cin is very non-linear. Nelson Pass's big simple MOSFET amps show rising THD in the top of the audio band, even though response magnitude is flat. (I accept that this may not be the worst sound around; may be quite benign.)

Grid/base/gate current always becomes an issue when you start optimizing everything.

 

[clintrubber]

[quote author="PRR"]> avoiding that 'base' current penalty.

There's always a grid current. [/quote]

Sure, fully agreed, the FETs 'solve' the DC-bias current of BJTs but do come with own consequences.

 

[Guest]

[quote author="recnsci"][quote author="Wavebourn"]
My turn #2: differential cascade using Tango pairs (FET+ BJT with resistor in emitter)
[/quote]

What's Tango pair?

[/quote]

A combination of FET and BJT in parallel, where BJT has a resistor in emitter. First time I used it in 800W/2Ohm amp to drive subwoofer. BJT with 2 Ohm in emitter used up to 3 A, before the beta-droop, FET took the rest vaporizing binding posts and a probe tip.

 

[Guest]

[quote author="clintrubber"][quote author="PRR"]> avoiding that 'base' current penalty.

There's always a grid current. [/quote]

Sure, fully agreed, the FETs 'solve' the DC-bias current of BJTs but do come with own consequences.[/quote]

Right. "All substances are poisons; there is none which is not a poison. The right dose differentiates a poison and a remedy." (C) Paracelsus

The topic is about the right dose of mixed poisons.