Interesting Amplifier Topology

[mawyatt]

Here's an amplifier with some unique properties that may be of interest to amplifier buffs, but first a little history, at least as best we can remember

Back in the 90s, the late Barrie Gilbert developed and interesting type mixer he called the MicroMixer. This mixer utilized the cross-coupled quad LO switching he made famous with his analog Multiplier/Mixer, however he utilized a very unique transconductor totally different than the ones seen in the earlier Multiplier/Mixer below the LO switching devices.

We're not sure if Barrie was the originator of this unique transconductor amplifier topology, and likely utilized before the MicroMixer. Whether used as a amplifier or mixer this circuit has some very unique and useful properties.

First off the amp is a single ending bi-polarity input with a pseudo-differential output, but operates from a single supply rail. This is accomplished by steering the input current between a pair of output transistors similar to a class-B stage but with two different current outputs. These currents are theoretically unlimited and only constrained by the transistors. The pseudo-differential type is because the output currents are only sunk, and when the input signal is sufficient amplitude, one side is completely cut-off similar to a class-B amp with the crossover current set by a steady state bias level, and the output current is "steered" to the active side inverted or non-inverted.

Another interesting circuit feature is the input impedance is symmetrically controlled about the steady state bias condition and plateaus with large input signal levels of either polarity. This property allows the control of harmonic generation and has been utilized to create harmonic nulls as the input power is swept.

We've included some notes created from memory, and utilized these with 200~300GHz SiGe bipolar devices in some of our custom IC work way back, so hopefully they are somewhat close. Also conjured up an LTspice circuit to revel some of the unique properties (just used 2N3904s tho, so obviously completely different devices & parasitics from what we employed).

Here's how it operates, at least a very simplistic view. The + input current signal causes additional current (above Ibias) to flow in Q1 causing it's collector potential to rise and begins to shut off Q3. Q1s current is mirrored by Q2 and becomes the + Output Current. As Q3 begins to cut off its' effect along with R3 on the input impedance diminishes and the input Z is now dictated by R2, dynamic 1/gm of Q2 and R1. In the other input current direction, Input current is pulled from Q1 collector and Q3 emitter until Q1 cuts off due to the voltage drop at Q1's base/collector, and Q3 supplies all the negative outgoing current thru R3 and it's collector sinks the inverted Io_bar Output Current. At large negative input currents, the result is symmetrical as the input Z approaches R3, the dynamic 1/gm of Q3 and R1.

The input impedance is tricky to revel in simulation, and requires the proper use of the derivative function {Zin = d(Vin)/d(Iin)} as well as removing the input offset with the offset bias source V1. Note this circuit is self biasing and an AC coupling capacitor would be used in series in actual use.

A quick analysis of the input shows that the input impedance at no signal level should be:

Zin = R1 + (R2 + 1/gm)/2, or ~ 57 ohm with value shown, and for simulation:

Zin = d(Vin)/V(Iin) as shown.

1/gm is (kT/q)/Ic or ~50.5 ohms at Ibias = 512ua.

With large input signal of either polarity then:

Zin approaches R1 + R2 (or R3) as 1/gm -> 0  for large currents, or 41 ohms.

Note that Q4 and R4 are optional in the notes, and not used in the LTspice simulations. The higher CE voltage for Q2 helps compensate for the finite beta of Q2.

We've used the DC sweep function (1VPP, 20VPP) from 50 ohm source to show the Input Z and how the Output Currents behave, note at zero input the outputs are both at Ibias. We also used a 20VPP sinewave input from 50 ohm source and you can see the differential output current and individual output currents.

The analysis of this topology quickly becomes transcendental for just about every important metric and requires some serious computing power and simulation software. We had the benefit of both back when employed but not so today after retirement, so very limited. A couple interesting topics wrt to this, was long ago recall a note about a Post Doctoral student being "employed" to run analysis and simulations to find the somewhat unique "Harmonic Nulls" of this topology, this required optimization of the bipolar device characteristics, resistor values and bias conditions, an enormous task indeed. One result that came out of this efforts was not only the means to achieve the nulls, but also the use of resistor-inductors, these being the series resistor with purposeful series ESL inductance. If you note the arrangement of Q1 and Q2 one might spot the use of the emitter degeneration inductance to achieve input matching popular with low noise RF amplifier designs, works well here also!!

The values we've included are from past memory and were for a specific result, so may be in error regarding optimizitation (especially wrt Harmonic Nulls). Speaking of the Nulls, these were simulated by sweeping in the input power levels and plotting the fundamental and 3rd harmonic as a function of input power level, recall a very time consuming process. The differential output currents can be utilized in various ways, one of which is with a differential transformer.

Anyway, hope some folks find this topology interesting, we certainly did a few decades ago

Best,

[magic]

In terms of dynamic input/output resistance, this input is equivalent to a standard class AB, push-pull, complementary emitter follower output stage - the sort of circuit used in most audio power amplifier outputs.

Its dynamic resistance is the parallel combination of dynamic resistances of the (symmetric) "push" and "pull" halves, each consisting of a series combination of a fixed resistor and a PN junction. Dynamic resistance of the PN junction decreases with load current, which helps to compensate for the complementary half going open circuit and effectively disappearing under class B conditions.

The problem of setting up standing bias for maximally flat dynamic resistance over load current has been studied thoroughly by audio amplifier designers. The trick is to ensure that under quiescent class A condition (zero load current / no input current to the TIA), each of the push/pull halves has twice the dynamic resistance as either half under heavy class B operation. If intrinsic emitter resistance under class B conditions can be neglected, then class B dynamic resistance is simply {R2}. Therefore, under quiescent condition, intrinsic emitter resistance of each transistor must equal {R2}, which is achieved when one thermal voltage exists across each of R2 and R3.

In audio amplifier practice, bias adjustment is usually empirical, for minimum loading induced crossover distortion. It is quite possible that a small signal circuit like this, if operated under a wide range of temperature, could benefit from constant gm biasing.

The resulting resistance vs loading can never be truly ideal as far as I understand, there is an inevitable increase in dynamic resistance under moderate class B conditions, but it is still fairly smooth. The simulated plots below show voltage gain versus load voltage of a complementary EF stage employed to drive heavy resistive loads.
http://douglas-self.com/ampins/dipa/dpafig15.gif
http://douglas-self.com/ampins/dipa/dipa.htm

The drop at the extremes of load current seen above is caused by large signal limitations of the transistors and only appears under heavy current loading.

[mawyatt]

Certainly audio Class-B & Class A amps in various flavors have been studied for awhile, however this topology is useful from ~DC well into the 10s of GHz region and remains well behaved, and operates as a single ended to pseudo-differential conversion from a unipolar supply while maintaining a respectable input Z to the input source (usually 50 ohms) over a large input bipolarity signal swing, and unlike a typical Class-B stage doesn't require complimentary devices! Things we don't see a typical audio amp capable of achieving, especially considering the simplicity of this topology without feedback and also allowing "insertion" into other useful functions such as Gilbert's original mixer use.

One area we haven't seen addressed often is the effect of signal level on the amp input impedance, something all amplifiers are subject too. This fundamentally creates signal distortion right at the amp input terminal as now the input Z is a function of the signal which modulates the actual amp input signal level wrt the source impedance! Making the input Z very high wrt the source impedance, operating with very low signal levels, or effectively padding the input, and/or all of these is the usual solution. This becomes more difficult with wide bandwidths, where one is usually regelated to the 50 ohm world, and different approaches are often required, which was our case.   

We employed a simple bias scheme to compensate for gm = Ic/(kT/q) with a bias that was proportional to temp (PTAT),   Ic = Ico(T/To), and gm = Ico(T/To)/(kT/q) or gm = Ico/(kTo/q), thus d(gm)/dT ~ 0, where To is a set-point constant (~300K).

Anyway, thanks for the reply and think if you spend some time with this topology as we did many decades ago, you'll begin to appreciate the inherent simplicity yet quite respectable performance with somewhat unique properties. This core topology was utilized in some demanding chip developments in both the amplifier and mixer configurations, much later the mixer was replaced with another unique but more complex structure with extremely high performance, but that's another subject if folks are interested.

Best, 

[moffy]

Thank you mawyatt, you always provide something interesting and fun! Makes quite a decent rectifier circuit, but I guess that the transconductance would be temperature dependent and that transistor matching would be important.

[mawyatt]

Thanks moffy, glad you enjoy some of these circuits and discussions!! Always believed Electronics should be fun and try to convey such, even when a Grad School Adjunct Prof long ago!!

The transconductance is easily temperature compensated by means of the PTAT bias previously mentioned, however the transistors should be matched reasonably well and in thermal unison (like on an IC) for best results, altho one can still get reasonable results with discrete devices on a Prototype-Board as shown below.

Slapped together some 2N3904s from the junk box (not matched) and a few 10 ohm resistors for {R2} in LTspice schematic (didn't have any 18 ohm ones) and a pair of 47 ohm (in parallel 23.5 ohms) for {R1}. Used a pair of 100 ohm resistors in the collectors to allow the currents to be "sensed", and 12V VCC supply. Here's the results with a 1KHz 5VPP Sinewave from a 50 ohm source, note C1 and C2 are DC Offset by 9V @ 10ma/div, Orange trace is C1-C2 @ 20ma/div, the circuit seems to work as described.

Best, 

[moffy]

Thanks moffy, glad you enjoy some of these circuits and discussions!! Always believed Electronics should be fun and try to convey such, even when a Grad School Adjunct Prof long ago!!

The transconductance is easily temperature compensated by means of the PTAT bias previously mentioned, however the transistors should be matched reasonably well and in thermal unison (like on an IC) for best results, altho one can still get reasonable results with discrete devices on a Prototype-Board as shown below.

Slapped together some 2N3904s from the junk box (not matched) and a few 10 ohm resistors for {R2} in LTspice schematic (didn't have any 18 ohm ones) and a pair of 47 ohm (in parallel 23.5 ohms) for {R1}. Used a pair of 100 ohm resistors in the collectors to allow the currents to be "sensed", and 12V VCC supply. Here's the results with a 1KHz 5VPP Sinewave from a 50 ohm source, note C1 and C2 are DC Offset by 9V @ 10ma/div, Orange trace is C1-C2 @ 20ma/div, the circuit seems to work as described.

Best,

Very cool! That looks just like the LTSpice sim.

[T3sl4co1l]

Interesting. Love the class AB crossover, that sure simplifies biasing.

Q1 is diode-strapped, so Q3 is just cascode of input current; this greatly limits power gain (current gain is unity), doesn't it?

Tim

[mawyatt]

Agree, so simple yet so elegant in operation.

Yes the current gain is ~ unity, kinda like a Single Ended To Differential Class-B Common Base Amp from an input current standpoint.

Best,

[mawyatt]

Very cool! That looks just like the LTSpice sim.

Didn't have convenient way to show the differential output currents other than the voltage across a resistor, so the currents appear inverted with zero current being at the display top. Note supply was 12V, so the display center is 9V due to DC Offset and the top is 9V + 3 divisions at 1V/div, or 12V as it should (this becomes 0 current ). Since 100 ohm resistors were used the scale factor is 1/100 (10ma), and at 1V div this becomes is 10ma/div. 12V VCC was used to keep the dissipation down as things begin to get warm as you pump ~100ma thru this circuit

Best,

[magic]

No doubt that this works but there is very obvious crossover distortion

You could try to re-bias it for 26mV across emitter resistors.

[mawyatt]

Agree some crossover distortion, however biasing at 2.6ma (26mv/10ohms) would be high and cause a significant input Z drop, and a redistribution of {R1} and {R2} values. With {R2} at 10 ohms as we have, this would cause {R1} to become ~40 ohms to maintain a zero bias ~50 ohms and possibly negate some advantages of this topology (which we've already done using 10 ohms for {R2} rather than 18 if memory serves correct). I mean one could just bias at 10ma or higher, eliminate {R2} and make {R1} ~50 ohms, but this is nothing more than effectively padding the input Z and making the core input Z ~ 0 ohms as a ~ virtual ground while raising the static DC power consumption, nothing wrong with this approach and may be useful in some applications.

Please remember this topology is not intended as "Audio Amps" and in wide-bandwidth systems much effort is placed in minimizing the odd order (mostly 3rd) distortion, which usually entails significant DC power consumption where the active signal currents are small relative to the bias currents (classic Class-A), and/or negative feedback. This is where we and others found significant benefit (aside from the simplicity) of this topology with the ability to design in "Harmonic Nulls", which are achieved by a delicate balance between the real impedances and the dynamic transconductance of the active devices.

Best,

[mawyatt]

Just a few quick plots at 1KHz, 10KHz, 100KHz and 1MHz, all at 5VPP input. 

Also simulated Input Z with measured values.

Here's the setup on a Photo-Board.

Best,

[moffy]

Interesting. Love the class AB crossover, that sure simplifies biasing.

Q1 is diode-strapped, so Q3 is just cascode of input current; this greatly limits power gain (current gain is unity), doesn't it?

Tim

Indeed it does look like class AB crossover. So I ran the sim and it looks very similar, but a little more complicated, thanks T3sl4co1.

[mawyatt]

Did a few FFTs with a 4VPP input at 1KHz (161), 10KHz (162) and 1MHz (163) before we disassemble the setup. Even with the incorrect {R2} (at least recalling from memory), the results aren't too bad considering the kludge setup and the junk bin components. Thinking a traditional DC Class-A amp swinging this level of input voltage and current (50 ohms input Z) from a single 12V supply wouldn't produce much better results without either series, shunt, or both negative feedback, and would be required to have >40ma DC bias and likely a higher supply (DC shunt feedback eats up supply range).

Anyway, hope some folks found this topology interesting and maybe useful, certainly fun the play around with utilizing junk components. Wish we could remember the actual values for the harmonic nulls, but those are beyond our limited memory and likely wouldn't be useful without detailed hi-current bipolar device models (MEXTRAM or HiCUM, VBIC isn't good enough hi-currentwise) for the devices utilized.

Best,