EZ-Dump: dump your current without really trying

Here is a new project, virtual for the time being, but I'll certainly give it a try one of these days.

It is a current dumping amplifier, but with an originality: the class A amp is based on a boosted-rails opamp. This means that the signal part of the amplifier is minimal, and that any standard opamp having a maximum supply of 36V (or even less) is suitable.

The pic shows the circuit and its operation, with the supply rails moving in a "telescopic" fashion about the output voltage, and keeping their difference ~constant.

The basic principles of the technique are detailed here: http://joebrown.org.uk/images/DualPSU/BootstrappingOpAmps.pdf

Unlike a conventional opamp amplifier, two resistive dividers are involved in the gain, and the actual gain is determined by the difference in the ratio's of the dividers.
If we call K- the ratio of the divider of the (-) output and K+ that of the (+) input, the gain can be written as (1 - K-)/(K- - K+).
 

Attachments

  • EZdump1.png
    EZdump1.png
    95.2 KB · Views: 2,250
  • AutoBoost.asc
    4.9 KB · Views: 142
Member
Joined 2011
Paid Member
Why use a superfast rectifier for D1 and D4? Why not BD139s / BD140s rigged as diodes? At least the errors in the SPICE model of the diode would track the errors in the SPICE model of the emitter follower. Looks like there used to be 1N4001 diodes in the circuit, why remove those?

Why connect C3 across the emitters of the bootstrappers? Why not connect it across the bases?
 
Why use a superfast rectifier for D1 and D4? Why not BD139s / BD140s rigged as diodes?
You have to remember that at this stage, the circuit is still virtual: I just wanted to bias the BD's properly without complication, but when a prototype is built, this kind of detail will be refined

Why connect C3 across the emitters of the bootstrappers? Why not connect it across the bases?
This is basically the reason:

A 220n should provide low impedance well into the VHF region - I wouldn't trust an EF with fT = 50 MHz transistors to do the same. You could add another one across the bases though.

At first sight, a capacitor across the bases looks more effective, as its value is multiplied by the Hfe of the transistors, but in fact, the opamp doesn't require a low frequency bypass: it is only at HF and VHF frequencies that instabilities due to an excessive supply impedance can show up.
The impedance seen from the emitter of an EF is not something clean and real like |0.026/Ie| at high frequencies, it much nastier than that, hence the local decoupling.

I wouldn't add a cap between the bases, at least not without base-stoppers: a low impedance path between the bases and the emitters could also favor instabilities.
 
Then why not insert base stoppers too, which add to the real part but not the imaginary part of the emitter impedance?
Simply because I do not see the need for it: with modern opamps (I mean post-1975) it will make practically no difference.
When the time comes to split the last ppm, we may come back to it, but at the moment, I don't find it necessary

Aren't you making an oscillator ?
Negative feedback 100k+100k and a posirive feedback 100k+91k+Rsource.
Ones the source is more than 9k I think oscillation starts.
Mona
Clever observation, there is indeed something there.
It will be adressed later.

OK, it's time to progress towards the real dumper: the current dumpers, Q1 & Q2 have been added, as well as the bridge components: R11, R12, L1, C6.
Some values have also been adapted.

Sout is a test probe scaling the output voltage by exactly the gain value, allowing comparisons with the input signal. This is necessary to fine-tune the bridge components, because artifacts are too small to be directly visible on the output signal.

Next pics show the 1K & 20K THD. Not too bad considering the simplicity and crap components used.

The next two pics show the error residue signal between Vin and Sout; even with the probe, it is necessary to zoom to see anything: the amplitude of the residual artefact is ~50µV

The opamp bootstrapping scheme hides some subtleties: Mona has spotted one, it is the input impedance: it is actually negative, and if the source impedance becomes too large, the feedback becomes positive.
Here, this issue is dodged by shunting the input with a resistor whose value is slightly lower than the negative input impedance: this eliminates all risks of positive feedback, and magically raises the resulting impedance.
The uncorrected impedance is -2K, and when paralleled with +1.8K, the result is +15K, as the input current demonstrates.
 

Attachments

  • EZdump.asc
    10.3 KB · Views: 127
  • EZdump7.png
    EZdump7.png
    98.3 KB · Views: 414
  • EZdump6.png
    EZdump6.png
    94.2 KB · Views: 365
  • EZdump5.png
    EZdump5.png
    97.4 KB · Views: 1,768
  • EZdump4.png
    EZdump4.png
    136.9 KB · Views: 1,838
  • EZdump3.png
    EZdump3.png
    130.2 KB · Views: 1,969
  • EZdump2.png
    EZdump2.png
    54.5 KB · Views: 2,103
Another issue arising from the interaction of the current dumping scheme with the bootstrapped opamp also needs to be fixed: the bridge component values are OK for the dumping section, but they aren't compatible with the normal operation of the VAS.

The first pic shows the frequency response: the -3dB point is at ~15KHz, which is clearly unacceptable. What's going on?

The problem is that the value of the reactive components of the dumping bridge are multiplied by the bootstrapping bridge gain, resulting in a heavy rolloff.

The theoretical solution is pretty obvious: the boostrapping bridge needs to be compensated. This will change nothing to the dumping characteristics, but it will correct the frequency response.
There are two problems with this approach though: unlike the capacitor, the inductor will be difficult to compensate using practical, realistic components. This means that the compensating cap will need to be much larger, because it will also have to compensate for the inductor, and this will exacerbate the second potential problem: in this case, compensation means tying a capacitor between the output of an amplifier and its (+) input, something I feel rather uncomfortable with... Looks like a recipe for an oscillator.

In sim, it works and it looks stable, but I wouldn't trust one second such a result without making a reality-check.
With 24pF, the frequency response can be extended (in theory) to ~36KHz, which is not large, but sufficient.

This will need to be tested in reality, but fortunately there are also safer solutions.
 

Attachments

  • EZdump8.png
    EZdump8.png
    129.6 KB · Views: 359
  • EZdump9.png
    EZdump9.png
    113 KB · Views: 323
I mentionned a better solution to the excessive HF rolloff: here it is.

It involves nesting the circuit into another, outer loop.

Using this technique, the frequency response can be easily flattened, but it also brings a number of additional advantages:
  • The overall gain can be freely chosen, without the akward constraints of the two dividers
  • The available loop gain becomes much larger, thus improving other performances markedly
  • The input is now a conventional, high-impedance (+) in, without the effect a negative resistance
Some adjustments to the gain of the initial circuit have been made in order to balance fairly the gain between the two blocks.

Thanks to the well-behavedness of the original amp and the gain structure, the compensation is very straightforward.

Of course, this solution requires one more opamp/channel, but considering the performance, the flexibility and the safety it brings, it is a reasonable price to be paid.

The THD@1K is now very much below the ppm level, and the THD@20K isn't bad either.
With good OP transistors and super-opamps instead the the MJE pair and the general-purpose LT1056, a stellar level of performance would be achievable
 

Attachments

  • EZ2dump.asc
    11.4 KB · Views: 92
  • EZdump12.png
    EZdump12.png
    141.8 KB · Views: 747
  • EZdump11.png
    EZdump11.png
    138 KB · Views: 761
  • EZdump10.png
    EZdump10.png
    118.4 KB · Views: 766
Hi Elvee
great performance, but switching hiss will maybe be an issue

Yes, it is still very much a work in progress, and therefore (easily) perfectible.

In this case, the bias network was a bit starved, and by slightly increasing the current and using 22µ bypasses, the higher harmonics are pushed ~110dB down wrt. the fundamental.
That part of the circuit is certainly not going to remain that way: as I said earlier, it is a simple and convenient way to bias the class A amp, but it is far from optimal.
In the end, I may use a diamond buffer instead of diodes: it will have the advantage of easing the load on the opamp.

BTW, with this fix the 1K THD becomes truly insignificant
 

Attachments

  • EZdump13.png
    EZdump13.png
    109.2 KB · Views: 736
  • EZdump14.png
    EZdump14.png
    187.7 KB · Views: 277
Here is a more practical, rationalized version: the main changes are the addition of a diamond buffer and capacitors C9 and C10

The diamond buffer is also used as a Vbe multiplier, allowing an easy adjustment of the class A amplifier's quiescent current with a single resistor, R25.

The capacitors establish a link between the opamp's supplies and the output of the amplifier, where the opamp's output current returns and has therefore to be referenced in some way. The capacitors provide a HF "shortcut", beneficial to the performances.
The capacitors need not be large, fortunately, and cannot be as they would disrupt the operation of the supply bootstrap.

Thanks to the improvements, the 1K THD has dropped to ~50ppb, and the 20K THD to 14ppm, almost pure second. Now, most of the higher harmonics are >120dB below the fundamental.
Small signal BW is >200KHz and FPW BW 80KHz.

I will probably build this version one of these days
 

Attachments

  • EZ2dumpVar.asc
    14 KB · Views: 105
  • EZdump18.png
    EZdump18.png
    80.6 KB · Views: 275
  • EZdump17.png
    EZdump17.png
    145 KB · Views: 227
  • EZdump16.png
    EZdump16.png
    144.4 KB · Views: 395
  • EZdump15.png
    EZdump15.png
    54.1 KB · Views: 427
I am a bit hesitant: this circuit exists only in theoretical form, and as you know, the real world generally requires adjustments, optimization and modifications to perform properly.

If you want to take the risk, you are free to do so, but then you have to be prepared to make some debugging work (unless we are extremely lucky).

Anyway, I shall certainly build it, but I don't know exactly when
 
Member
Joined 2014
Paid Member
Elvee, a few practical questions:
1) for all small npn/pnp: ksa1845/992 ?
2) for drivers: ksa3503/1381 ?
3) for output: MJL21194/21193 ?
4) is L1 an air coil? 1mm wire ?
5) R24 2W ?
6) Any other resistors need to be bigger than 0.25W ?
7) is 4Ohm load OK?
Any other practical considerations?