JOB circuit by GOLDMUND

[Kal]

Hello,

I recently read a review of a power amp by GOLDMUND that uses the JOB-circuit. This amp seems to be breathtakenly fast... and great sounding!!!
And it is only made with a very few parts...

Has anyone more information about the JOB-schematic/circuit except for what is available at www.goldmund.com or www.jobsys.com?

Thanx

 

[lgreen]

Good Question

I asked the same thing awhile ago, see my earlier thread:

http://www.diyaudio.com/forums/showthread.php?postid=189965#post189965

I'm with you, I'd like to make one of these if provided with the right information on what's in there.

 

[arthur]

difficult to copy
it contains a double sided pcb with lots of smd transistors
no markings on the transistors also

 

[subwo1]

double current mirrors

So far, the fastest topology I have found is current mirrors connected to the collectors of dual input pairs. They provide active base pull-down for the upper and lower VAS transistors. I wonder if that topology is what they use. Simulations I have done indicate a slew rate up to 200uV/s when used with complementary source follower mosfets. In the simulation, output device idle current was set at 500ma. Maybe these ideas will give you some guidance.

 

[Elso Kwak]

Re: double current mirrors

So far, the fastest topology I have found is current mirrors connected to the collectors of dual input pairs. They provide active base pull-down for the upper and lower VAS transistors. I wonder if that topology is what they use. Simulations I have done indicate a slew rate up to 200uV/s when used with complementary source follower mosfets. In the simulation, output device idle current was set at 500ma. Maybe these ideas will give you some guidance.

Hi, I would replace the VAS transistors by Darlington connected transistors. The current mirror likes to see a high impedance as load.

😎

 

[subwo1]

Hi Elso, it would just happen that we end up with such a trade-off. The fast, active pull-down of the current mirror likes the high impedance of a Darlington pair with slow response. But then several resistors would be needed to make up for the forgone self-biasing nature of the current mirror. Maybe lets just make that waif of a current mirror sweat and drive that single transistor.😉

 

[Charles Hansen]

Subwo1, try simulating the circuit with the Hitachi lateral MOSFETs as output devices. These are what are used in the Goldmund. Their lower input capacitance may speed things up even more.

Also, you could try adding separate emitter followers between the current mirrors and the VAS. I think this is what Elso meant, and it should also speed things up.

 

[Elso Kwak]

Darlington

Also, you could try adding separate emitter followers between the current mirrors and the VAS. I think this is what Elso meant, and it should also speed things up.

Hi Charles,
No I meant something like the VAS stage as here:
http://home.kimo.com.tw/skychutw/ampzilla/index.html
Unfortunately the AmpzillaII schematic is gone but here is a detail.

 

[Charles Hansen]

I think the Ampzilla circuit is a variation on what Borbely called the "Lender circuit", where the emitter of the VAS is driven by one phase of the diff pair, while the other phase drives the base of the VAS. I can't see where the collector of what looks like an emitter follower is connected.

At any rate, adding an emitter follower to drive the base of the VAS will eliminate the slowdown from the Miller capacitance of the VAS. This is what I think we were both suggesting to Subwo1. However, I don't think it is possible to combine the "Lender circuit" with the current mirror.

 

[Kal]

Hi Charles,

I really like our amps, too!!! One of the best amps I ever heard!
But unfortunately I can not afford them at the moment...
I really would like to have your AX-7 in a unbalanced version(only RCA inputs).
For some reason I prefer unbalanced interconnects...

 

[subwo1]

Hi Charles,
I simulated the current mirror circuit with generic hexfet models and 2sk351 and 2sj102 lateral (?) ones as those were the ones available. The slew rate with the hexfets came out as 210v/us while the laterals gave 165v/us. The output idle was set around 500ma. The real life results would be somewhat different.

What I see with the Ampzilla II circuit is the feedback operating with a folded cascode in regard to the VAS emitter drive; I call it that term anyway.🙂 In reality, though the current mirror I show is elegantly simple, some value of VAS emitter resistors of at least the low single digits is really needed.

I tend to think that adding emitter followers before the bases of the VAS will be worse since the active pull-down will be lost. I plan to try it out on the simulator for a look-see. I will also need to pad the emitters of the current mirrors with diode junctions or resistors, it seems, to shift the voltage levels away from the rails to compensate for the diode junctions of the emitter followers.


Hi Elso,
I think I will pull back up one of the folded cascode feedback circuits on the simulator and see if I can try out that emitter follower buffer for the VAS base. I think the Ampzilla II circuit may become faster if the VAS base pull-down resistor is reduced by around a factor of 10. But the question is where does that emitter follower collector go. It appears to go toward the location of the constant current source which is sorta strange.

 

[Charles Hansen]

I simulated the current mirror circuit with generic hexfet models and 2sk351 and 2sj102 lateral (?) ones as those were the ones available. The slew rate with the hexfets came out as 210v/us while the laterals gave 165v/us. The output idle was set around 500ma. The real life results would be somewhat different.

Yes, the real life results are always different.... 🙂

In real life you wouldn't use just one pair of lateral MOSFETs for outputs (at least if you wanted to do it right). Instead you would use at least 2 pairs (and maybe as many as 8 pairs) to replace the vertical MOSFETs.

Now on first glance that would seem to make things worse, as the input capacitance would be increased by adding more devices. But in actuality things get better. This is because the (relatively large) Cgs has some effect with a single pair of devices, due to the low transconductance of the lateral parts. In other words, the "source" doesn't "follow" the gate very well with only a single pair driving 8 ohms.

You could spend a lot of time adding devices in your model, but one quick thing to try is to increase the value of the simulated load from 8 ohms to (say) 32 ohms and see what happens to the slew rate. Let me know what you find out.

 

[subwo1]

Ok, with the load resistor 32 ohms it came up with 240v/us for the lateral mosfets. That increase is 50%.

 

[Jan Dupont]

Elso Kwak

You can find the Ampzilla II schematics on my homepage under Schematics 😉

 

[Graham Maynard]

Hi subwo,

I see your circuit is maturing. ( Wondered what was happening !)

I note your use of the dual NFB networks to maintain a lower stabilising impedance at radio frequencies.

I agree with you about the mirror pulldown, because mirrored differentials can swing through reasonable currents without becoming the most significant source of distortion.

If you added an emitter follower you would need approx 100 ohms at the VAS base-emitter for equivalent speed.
The follower would also reduce closed loop distortion by a factor of say 30, but maybe cause instability, and might necessitate separate resistance sensed current limiting at the VAS emitter.

The input filter catches my eye as having a earcatching turnover of less than 10kHz !

Your circuit looks so simple, are you going to build and audition it for us ?

Cheers ............ Graham.

 

[slowhands]

Re: double current mirrors

So far, the fastest topology I have found is current mirrors connected to the collectors of dual input pairs. They provide active base pull-down for the upper and lower VAS transistors. I wonder if that topology is what they use. Simulations I have done indicate a slew rate up to 200uV/s when used with complementary source follower mosfets. In the simulation, output device idle current was set at 500ma. Maybe these ideas will give you some guidance.

On paper this looks great, and the mirrors will lower distortion dramatically. But be very careful with physical layout. You can simulate this all day long, but if the physical layout is not really good, it's an oscillator.

I ended up going to shielded leads for the input line and gate drive, with ferrite beads on the gate, because the output stage is very apt to oscillate up at 10MHz. Once it starts, you can't keep it out of the rest of the amp.

 

[Charles Hansen]

Ok, with the load resistor 32 ohms it came up with 240v/us for the lateral mosfets. That increase is 50%.

OK, that confirms my suspicions. The trick now would be to put 3 or 4 pairs of lateral parts in parallel and see how much of the speed increase is retained.

By the way, part of the slew rate from the original circuit using the IR output devices isn't real. This is because 1 ohm gate resistors aren't enough to ensure stability. I think a real minimum would be more like 10 ohms, but I would suggest more like 50 ohms for a real amplifier.

 

[eLarson]

I couldn't let this, er, pass.

Try putting ~100-200 ohm resistors at the gates of all your MOSFETs, and as close to the MOSFETs as possible.

That may help you some re: oscillation.

Erik

 

[sam9]

Going back to the Goldmund product, I finally found some info on their site besides assertions about how great they are. (I'm sure the sound just fine, there aren't that many total dogs even in Circuit City.) Their FFT analysis is kind of interesting - lots of high order harmonics. 2nd harmonics definately do not dominate. Once power ratings were the big deal regarless of other characteristics, then THD or THD+N became the "hot button" for marketing departments. Damping factor got a lot of ad copy for a while. It seems like slew rate is the top techno-babble now.

When any one spec gets pushed for marketing reasons I suspect other important characteristics suffer. They certainly did when some of the earlier numbers were pushed past what was reasonably required.

I guess I'm just grumpy today. Some terms bug me. One is "fast" when applied to audio. To me the only "blindingly fast" amps are the ones in the dash of Porsches, Ferraris etc blasting doen the Autobahn.

 

[subwo1]

Hi Graham,

glad you found me again. Yes, the current mirrors were a great advancement. The dual feedback networks provide a high frequency boost and phase compensation, for sure. I detected a 25ns difference between the input and output singals. I was thinking the same thing about emitter followers potentially adding more phase shift and instability. I determined the filter turnover by approximation, maybe wrongly based on the RC time constant, to be near 100khz. I unfortunately cannot determine a construction date.:-(

Best regards

Hi ACD,

Now I see that the collectors of emitter followers driving the bases of the VAS pair on the Ampzilla II draw current from the tails of the opposite differential pair through the voltage sources which are the emitters of their constant current sources. Huh? 🙂

Take care