The purpose of this work is to compare the performances of current feedback amplifier topology to voltage feedback one.
We have chosen a very simple, but amazingly good amplifier -it use only 6 transistors- from a very smart audio designer: Andrej Lackner (aka LazyCat).
We have tried to make from-it the best and closer
version we could in VFA topology.
That means: No modification of the VAS & power stage, same currents in the input transistors and same gain in the both input stages in order to compare only the bahavior of the two topologies everything equal elsewhere.
In all following images, the CFA's curves are green, the VFA are blue (or red).
VFA input stage design
The first work was to replace the input stage by a traditional long tailed pair. It requires the addition of two extra transistors.
We need to make the currents, the collector resistance in each transistor and the output DC voltage and AC gain the same it was in CFA, in order to get the following stages totally unchanged for a fair comparison.
Setting the amplifiers.
The current sources, powering the input stages, had been precisely adapted to power the two sides of the VFA, keeping the currents equal for each transistor in both topologies while keeping a near zero DC offset in the output.
Let us disconnect the feedback resistances from the amp out (replaced by ground) and tune the current sources for DC ofset and quiexcent current, while emitter resistances of the VFA are adjusted in order to get an open loop gain equal to the one of the CFA. All that is done at 100Hz to be far enough from HF & LF cutoff frequencies.
It was a little tricky and acrobatic, but the result is there:
Output signals for 0.3mv input at 100Hz, open loop.
Time, now, to reconnect the loop and to measure the output stage's quiescent current...and here are the results:
|Amp||Q13 current||Q14 current||DC Offset||Quiescent current||Openloop gain nb||Openloop gain dB|
Close enough ?
Compensating the amplifiers
The first need is to optimize the two amplifiers bandwidths. Miller capacitances C5 & C7 had been added to both amplifiers, and carefully adjusted (0.5db full scale
) for the most extended flat bandwidth as possible, at the overshoot limit.
Then, a little capacitance had been added in parallel with the feedback resistances R5, R6 (R13 for VFA) to get the fastest little signals square wave response possible without overshoot (50kHz).
Here the resulting schematics from this fastidious work:
One of the main advantages of the CFA topology is that the emitter of the input stage, where is applied the feedback, presents a lower parasitic capacitance than the base of the inverting input of the LTP , where the feedback comes in a VFA.
By habit, for various and justified reasons, including DC offset, the feedback resistance is set the same than the base to ground one in the none inverting input.
This limit considerably the bandwidth of the input stage, witch, contrary of what a lot of people believe, is often, with modern fast power mosfets, the main node limiting the performance. The feedback resistance impedance makes a low pass filter with the parasitic capacitance of the base. We can see their effects here, where the golden trace is with a 10K feedback value and the blue one, the 1K.
The bandwidth and phases
Green is CFA while blue if VFA.
The open loop bandwidth
Well, is the extended bandwitch of the CFA due to open loop, or feedback, or both of them ?
Surprise ! The extended bandwitch is yet there. Oh, yes, this miller cap ?
Let's, just for this test, to make sure and to confirm this, to try the bandwidth with the miller cap of CFA equal of the VFA one.
We can see the response is more than positive: the CFA has now, with the same miller cap than the VFA, a less extended bandwitch than the VFA.
This means that the impedance of the LTP input stage is slightly lower than the CFA one.
Okay, now we know the miller cap influence (in our example) on the bandwidth, let's reset the miller of the CFA back to its good value.
Why don't have a look to the large signals square waves ?
What a difference !!! A little too much to be explained by the miller cap, i presume...Let us try the two amps with the same bandwidth, setting for a while the miller cap of the CFA = 22pf. This makes the two bandwidth equal. And, surprise... The large signals square waves are.. near unchanged. The same difference remains.
...let us measure the slew rate to put numbers on it.
The Slew rates
They had been measured, applying a +-2V square signal to the inputs of boths amplifiers (more than enough to saturate them). 10% of the top and bottom output signals had been removed, as habit, to measure the time, then calculate the slew rates.
Warning: As the models of the power devices used by Andrej were not available, the compensations he made on the gates of the power Lateral fets are not accurate with our used models. But similar in the two versions. I am sure that, in the real version of the VSSA, there is not such a (relative) speed reduction after +18V.
As expected, the CFA is here 5 time faster than the VFA. And this is not due to the miller cap, but the topology: the way the feedback signals are substrated from the input ones.
This is due to the two input stage topologies transconductances: The Long-tail Pair is "compressive" while the CFA is "expansive" (© Richard Marsh).
An other way to describe-it is the "Current on demand" behavior of CFAs. So well described in various paper, no need to pontificate on this here.
For 1V pp in input at 1KHz:
Fore better readability, the VFA distortions are in red in those images.
Surprisingly, the CFA version show less distortion than the VFA ! It seems to happen often when you try to push a vfa too fast ;-)
Notice: where is the supposed superiority of the LTP to cancel the even harmonics ? Distortions are equal at 100Hz (0.0033%). Yet at 1KHz, the phases problems seem to enter in action and the cfa is widening the gap when the frequency increase.
Not to forget that, as in sims, the devices are perfectly matched, a hard thing to do in real life, the LPT even harmonics cancelation is suposed to be at its best.
Now, just a simple image to show the IMD (intermodulation distortion, 20kHz+19kHz)
are quite similar in both topologies: IMD is not an input stage issue.
Same scale, is the VFA is sligly better ? Well, i'm not sure ;-). Another urban legend dies here.
Ok, thanks to KeanToken
from diyaudio.com, a more accurate simulation of the expected noise:
Nothing surprizing: two transistors adding their noises (VFA) are more noisy than one (CFA).
The power supply ripple rejection
We have applied a 0.1V signal to each power source, inphase and in opposition of phase. Here are the results:
Here, and as expected too, the VFA shows an huge superiority: 6db in differential mode, 18dB in common mode. CFA amplifiers need great care to use low noise power supplies, and good filtration.
Notice that this superiority decrease with frequency and CFA win at 1MHz.
A table to resume the numbers we got in this comparizon:
|Amp||Open loop bandwidth||closed Loop Bandwidth||Slew rate||Harmonic distortion at 1KHz ||Harmonic Distortion at 10KHz||PSRR at 1Khz mode common||PSRR at 1MHz mode commun||PSRR at 1Khz mode differential||PSRR at 1MHz mode differential|
|CFA||2.16 KHZ MHz||5.537 MHz||317V/µs||0.000836%||0.005612%||67dB||32.2dB||91.4dB||53dB|
|VFA||1.157 Khz||4.413 MHz||52V/µs||0.000978%||0.00771%||85dB||31dB||95.8||52.8dB|
Time, now, to work to improve the both amps, distortion side, and PRSS rejection for CFA. It is obvious, for the objectivists, that both amps will behave quite identical in audio banwitdths, and they often prefer VFA for PRSS performance.
Not my personnal (and subjective) point of view based on many listening experiences, as it seems slew rate is an important point for micro dynamic and "details" in transients reproduction.
Not to forget that, if we use the two extra transistors of the VSA LTP to build a little 20mA cap multiplier in order to filter the input & VAS stages power rails, this time, the CFA wins on all PSSR !
Thanks to Andrej for giving us the autorization to use his wonderfull VSSA. Amazing, despite its simplicity.
Thanks Dadod for the models and base VSSA sim schematic.
Thanks Richard Marsh, for too much things to be listed here.
Thanks to KeanToken for his noise simulation.
Here the sims and models you can download to play with and verify my numbers.
Please, don't believe too much in sims. Build and listen in real life.
But here, simulator is a good tool to enlight the differences. Even if the numbers are not reflecting 100% the absolute values in real life, it make sense in a relative comparizon.