Towards a wideband non switching Auto Bias power amp

Thanks again Ed. Any is feedback much appreciated.

For those who are interested in floating power supply amps, here's a brief tutorial post for my auto-bias amps. The LTspice run file is attached:
Floating-PS-cct.png

This is a simplified circuit that you can run to see what's happening and how it works. The power transistors have infinite current gain so a driver stage can be omitted. The autobias loop is also omitted and there is a fixed idle current of around 180mA.
Capacitance "Cbe" provides simplified HF rolloff.
The load is effectively floating but loads "Out1" terminal can be connected to ground using the link.
The supply rails "Pos and "Neg" are floating in this amplifier.
The load "Out2" terminal connects to C1 and C2 which are effectively output capacitors (they provide DC speaker protection).
The input signal is connected to the amps "Out2" node (which is normally grounded but doesn't have to be grounded to operate). This means the power transistors operate in Common Emitter mode -- which is also the transconductance or current outout mode.

Some voltage feedback is then added by sensing the load voltage at "Out2"

Further, some voltage feedback is also added by sensing the rail voltages "Pos" and "Neg" with equal value resistors R1,R2 so negative feedback pulls "Out2" voltage to half of the rail-to-rail voltage V2, giving effectively +/-21V across output caps C1 and C2. Feedback resistor R6 is only for AC feedback while R1, R2 provide both DC and AC feedback. The effective feedback resistance of R1//R2 is 50k and this is in parallel with R6 giving about 25k for feedback. With a 1k input resistor the voltage gain is about 25 times. The actual gain is more like 20 due to finite driver stage input resistance (not modelled here).

The output resistance (and DF) with this voltage feedback depends on the transconductance of the output stage. If the Gm is 5A/V then the open loop voltage gain is Avol= 5A/V x 8 Ohms =40. If the closed loop gain is 20 then there is only a factor of 2 of voltage gain left for lowering the output resistance, giving only about 6 dB of feedback and an output resistance of around 8 ohms. (I measured about 12 ohms with 0.2 ohm emitter resistors). This amp is intended for current drive of loudspeakers. For conventional speakers you can add a Sokol tuneable equaliser to add the needed speaker damping (Post 219).

You may wonder why I have the Gnd jumper option. It shows you can completely float the entire amplifier and PS when you feed the input signal to "Out1" and float the signal source. This can allow a floating bridge:
Floating-PS-bridge-cct.png

The demo file is attached. The "Out1" node can be grounded giving a grounded output bridge and a non-floating input signal for the LHS. The RHS must use a floating inverted input source, which could be a transformer or a second floating DAC.

In fact you are free to ground whatever node you would like - for example you could use a conventional +/-21V non-floating power supply (to feed other channels) as long as the the two signal sources are floating.

Even more wiered, you can hang more bridged amps off the same floating power supply - as long as the two signal sources are floating and all the speakers remain floating (just include an antifloat ground resistor somewhere in your sim to keep SPICE happy). In practice you may prefer to use a conventional +/-21V non-floating power supply for a ground.

Apologies if this is too brief and "a bit like drinking out of a firehose" as my friend Jan Didden often says of my tech ramblings. Maybe someone can convert it into ordinaryeese?
 

Attachments

  • Floating-PS-Autobias.zip
    1.6 KB · Views: 14
  • Floating-PS-bridge.zip
    2 KB · Views: 13
  • Like
Reactions: 1 user
You and Marcel have successful auto-bias circuits. Fast auto-bias avoids the problem of the bias being goofed up by some signal in the (possibly distant) past.

I think the floating supply is not inherent to the auto-bias circuit. Floating supplies can pass noise from the AC line. You may want to develop a conventionally-powered version.

ETA: You may want to clamp Q5 and Q6 to prevent saturation.
Ed
 
Last edited:
Hi Ed,

Thanks.
...Floating supplies can pass noise from the AC line.
Yes, the use of a grounded screen between primary and secondary can absorb most of the noise from the AC primary.

With my autobias circuits there is also the option of a floating signal source with a grounded power supply.

There is also Post 160 option using a high voltage opamp and a grounded power supply.

Cheers, IanH
 
  • Like
Reactions: 1 user
This post is similar to Post 220 but using UF54001 3A diodes.
Autobias-UF54001-MJL3281-slice-0v9-pcb.jpg

Autobias-UF54001-MJL3281-slice-0v9-cct.png

The model for the UF54001 is close to the MUR1615 TO-220 diodes and gives a similar XY plot to that shown in Post 220. Two UF54001's are in parallel mounted on copper tabs so they fit in the TO-220 pads of my first PCB for testing. In Post 220 the two diodes in the MUR1615 are in parallel so the electrical properties are similar to two UF54001's in parallel. I used 50m Ohm resistors in series with the diodes as shown. There is another 10m Ohms in the wiring an copper tracks, giving 60 m Ohms total. The idle current without the bias offset resistors (R2,R22) is 540mA and with 3k2 for R2 and a trimpot for R22 the idle current is reduced to 250mA which is the minimum idle current for best linearity with 60m Ohm series resistance.

Using music set for just below clip (3Vpk in and 42Vpk into an 8 Ohm speaker) the temperature of the diodes remains almost constant at 43C (28C rise) and the main heatsink 55C (40C rise) at idle and 58C with music. At the instant of stopping long term near clip music the idle current is 260mA and falls to long term value of 250mA withing 2 seconds. So the thermal properties on this PCB appear to be good enough for music use. The distortion according to simulations is in the 0.1% range for the first watt -- but distortion was not obvious with music for any power level even up to almost clip.
 

Attachments

  • Autobias-UF54001-MJL3281-slice-0v9.zip
    8.3 KB · Views: 19
Last edited:
  • Thank You
  • Like
Reactions: 1 users
Hi Pawel,

Good question. I assume you read my tutorial (Post 222) where it says:
...some voltage feedback is also added by sensing the rail voltages "Pos" and "Neg" with equal value resistors R1,R2 so negative feedback pulls "Out2" voltage to half of the rail-to-rail voltage V2, giving effectively +/-21V across output caps C1 and C2. Feedback resistor R6 is only for AC feedback while R1, R2 provide both DC and AC feedback. The effective feedback resistance of R1//R2 is 50k and this is in parallel with R6 giving about 25k for feedback. With a 1k input resistor the voltage gain is about 25 times. The actual gain is more like 20 due to finite driver stage input resistance (not modelled here).

The output resistance (and DF) with this voltage feedback depends on the transconductance of the output stage. If the Gm is 5A/V then the open loop voltage gain is Avol= 5A/V x 8 Ohms =40. If the closed loop gain is 20 then there is only a factor of 2 of voltage gain left for lowering the output resistance, giving only about 6 dB of feedback and an output resistance of around 8 ohms. (I measured about 12 ohms with 0.2 ohm emitter resistors). This amp is intended for current drive of loudspeakers. For conventional speakers you can add a Sokol tuneable equaliser to add the needed speaker damping (Post 219).
So R3 and R19 are for centering the output voltage to the output capacitors and in the process gives some global AC feedback. Then I added R6 which only adds AC feedback - my intention was to select a value that gives just enough voltage gain to suit my signal source, at present a PC sound card so the amp just clips at the peaks, which seems to be 3V peaks from my laptop. With the amp above (Post 226) I needed more voltage gain so I removed R6 altogether. The gain increased from 12 to 15 into 8 ohms. BTW it was a gain of 12 with R6 at 47k (not a gain of 20 as in my tutorial) because I had replaced R3 and R19 with 47k's (not 100k's) - I had to use 47k's because it was the closest I had in 0603 (since the PCB I had outsourced had several incorrect values).

But still with a the higher gain of 15 I still didn't get it to clip. With music the rail voltage is higher compared to the full power sinewave test, requires even more gain to just clip. Next I'll may lifting the gain to about 20 by shunting the input resistor R12 to maybe 800 ohms.

As for feedback for reduction in output impedance, I am happy with about 6dB of feedback for about 12 ohms output impedance. I consider the amps distortion is low enough in the first watt region with this amount of feedback, and the relatively high output resistance (by modern standards) will reduce speaker distortion (as you are aware from Esa Merilainen's papers) and the addition of a Sokol equaliser will add the needed speaker damping. To drive 4 Ohm Maggies, use two slices in parallel. And, at a guess, Maggies don't need an equaliser to add damping if used with this relatively high output impedance amp. How convenient;).
 
Hi Ian,
Here are some links:

Magnepan Product Manuals

Magnepan Tweaks (scroll down to reference schematics)

I wrote an article on how the MG1.6 works:

Modeling the MG1.6

Maggies are either 2-way or 3-way. The drivers are nearly perfect resistors, but the crossovers cause considerable impedance variations. The current models (MG1.7i and MG3.7) use series crossovers whereas past models used parallel crossovers. All are designed to have 4-ohm impedance and ~83dB/W efficiency. Maggie owners invariably have big amplifiers due to the high currents required (~10A peak).
Regards,
Ed
 
Last edited:
Hi Pawel - The MG1.6's crossover design is such that any series resistance will boost the midrange. Depending on the amount of resistance, the sound may range from subjectively more "musical" to "the speaker is playing underwater!"

The MG1.7i has a completely different crossover and will behave differently.
Regards,
Ed
 
Last edited:
Hi Ed,

Thanks for all that info on Magnepan models. It shows we can't get any good response from Magnepans when they are driven by high impednace amps (transconductance amps). So I stand corrected. And TS models don't seem to be needed. Instead a dipole radiator model, like yours, is needed. Nice work!

Attached are the equivalent circuits for the MG1.5 and MG1.6. This is just the current division into the resistive panel elements and not the acoustic levels over frequency. But it can show the relative shift when driven with a high impedance amp. I simulate Rout of 10 ohms. There appears to be something like 6dB deviation in the 200Hz to 2kHz region (after trimming the amps sensitivity to allow for gain loss with a high Rout).

As you suggested, the way that ensures closest frequency response with high impedance (transconductance) amps is to use an active system - 2 amps per louspeaker and an electronic crossover. It would avoid changing the capacitors and inductors which is not easy noe cheap.
 

Attachments

  • Magnepan-1v5.zip
    1.3 KB · Views: 14
  • Magnepan-1v6.zip
    1.3 KB · Views: 13
Here I run two Autobias amps off the same floating power supply as mentioned in Post 222. Good news!
Autobias-UF54001-MJL3281-2-slice-0v9-pcbs.jpg

One board (RHS) has it's "Out1" grounded as usual (see post 222). The other board (LHS) has it's input isolated using a 1:1 transformer. Both boards share the same rail capacitor and share the same AC output capacitors. But note -- the two boards have independent isolated/floating +/-9V supplies!

And it works as per simulations above:cool:. It allows left and right channels to run (almost) independently as with standard designs that share common power supply rails.

It also allows bridging - just by reversing the transformer primary phasing. With each channel independently I got 34Vrms into 8 ohms at the start of clip. Then with a bridge I got 63Vrms into 16 ohms. Rail voltages at clip were 110V for a single channel and 99V for the bridge. So that's twice the output swing as expected with a bridge. (But to drive 8 Ohms with a bridge with this rail voltage each side needs two parallel slices (boards). But that will have to wait till I uprate the transformer and heatsinking).

So, as long as each amp has its own isolated +/-9V supplies and the inputs of the amps are isolated, then two or more channels can run off the same floating power supply. And as stated in the tutorial you are free to ground the power supply, such as the center point to reduce noise from mains getting through to the amps - and again, provided all these floating amps have isolated inputs and isolated +/-9V supplies.
 
  • Like
Reactions: 1 user
Hi Ed,

Thanks for all that info on Magnepan models. It shows we can't get any good response from Magnepans when they are driven by high impednace amps (transconductance amps). So I stand corrected. And TS models don't seem to be needed. Instead a dipole radiator model, like yours, is needed. Nice work!

Attached are the equivalent circuits for the MG1.5 and MG1.6. This is just the current division into the resistive panel elements and not the acoustic levels over frequency. But it can show the relative shift when driven with a high impedance amp. I simulate Rout of 10 ohms. There appears to be something like 6dB deviation in the 200Hz to 2kHz region (after trimming the amps sensitivity to allow for gain loss with a high Rout).

As you suggested, the way that ensures closest frequency response with high impedance (transconductance) amps is to use an active system - 2 amps per louspeaker and an electronic crossover. It would avoid changing the capacitors and inductors which is not easy noe cheap.
thank you Ian,
I confirm, the sound remains high peaking at crossover frequency, maybe it soudns like from underwater, however this transimpedance sound smoothness makes listening pleasure...
regards
Pawel
 
Hi Ed,

But I notice the Yamaha HCA circuit (Post 18 in the PDF download) does not sense each side of the output stage currents only the output voltage. That would seem to be quite different to other Autobias circuits like in my Post 156 . So the HCA output stage would appear to be not a nsb (non-switching output stage).

My sims show the E Van Drecht patent (see Post 156) is also not nsb - but if you break the resistor node of R17,R18 connecting to the output then you can get nsb - that was one of my "eureka" moments that had escaped me for a long-long time! I think also peufeu's circuit (see Post 156) is not nsb - it would benefit from a 33 Ohm resistor across the two Schottky's to make it nsb.

My XY scope plots of the Autobias fed via the two emitter of the bias transistors, have shown that without the resistor across the diodes to make it nsb, that a glitch sometimes appears near clipping as one side turns off completely and has to restart leaving clip - which is very undesirable and audible. So it is vital for these versions of autobias. My sims without this keep on resistor across the diodes also show this glitch. My sims previous to Post 202 did not use this resistor and would sometimes show this glitch in the gm plots. This type of glitch does not happen in the convention biasing arrangement, only some autobias types.

Cheers, IanH
 
Hi Ed,

But Post 77 sim of that circuit in Post 75 says it is just "Class-AB with switching".

I haven't simulated that exact circuit, but looking at the sensing of emitter resistor voltage rather than a log diode voltage as in my circuits, you won't get the log-antlog conversion in the bias loop (there is just the antilog of Q15,Q44).

So it is fundamentally different to mine and the E Van Drecht patent and peufeu's including the LT1166.