AX100 100W Aleph-X Monoblocks

[macka]

Originally posted by gl
Well, to be totally honest one amp has the 15 ohm Source resistors and one doesn't. I'm still messing around with that detail.

Graeme

Graeme,

I think I have one channel up and running on a test jigged output stage with 4 240s biased to 1 amp.

About 5 volts offset at start up an 7.5 mv relative offset.

The other channel has a fault with relative offset at o.6 volts...could be anything and I will investigate tomorrow.

So initally it looks promising.

I will let you know how I progress.

Ian

[macka]

Both channels going now.

i'm really surprised by the low absolute and relative offset.

I should thank Erno.

No source resisters at this stage.


Ian

[EUVL]

Just enjoy the music.




Patrick

[gl]

Hi Dave,

It's great to hear that your system is running well. I truly wish I could come over and hear it.

I chose the bipolars for cascodes because I have lots of them, they're cheap, and I don't believe that JFET's do any better in this role. I jealously hoard my JFET's saving them for more significant roles.


Hi Ian,

Congrats on getting both channels up and running and on the excellent operating specs. Now for the big questions; have you done any listening and if so how do they sound?

Graeme

[macka]

Hi Graeme,

Not yet.

I will do the final build over the next two days.

Ian

[macka]

Hi Graeme and William and Patrick,

Its all working I just need to refine the physical layout of my Jfet modules.

I never graphed the Transconductance curves but wonder if sme variation in bias would make any difference.

Some changes.

I diced my 4,7 uf caps...those auricaps of mine sucked in that application. The offset is about 30 mv with the X2.5 preamp

The supply rails are 23 volts.

Subjectively I like it. Definately an improvement on the 9610 input device tonally. Its has similar but not identical voicing to the X250.5 which I guess is a good thing!


Ian

[EUVL]

> I never graphed the Transconductance curves but wonder if sme variation in bias would make any difference.

A typical Id vs Vgs.
Of course dI/dV = Yfs.

Patrick

[macka]

http://www.diyaudio.com/forums/showt...65#post1166765

Hi Patrick ,

I have been reading Williams very interesting Aleph J thread :
http://www.diyaudio.com/forums/showt...65#post1166765

and your excellent post above regards using the 2SJ74.

The mental gymnastics of this discussion are for the moment a lot to get one’s head around.

If I ready your comments correctly there are several considerations.

The trade off of open loop gain versus open bandwidth of the first stage and the dependency of the second stage open loop bandwidth on the signal source resistance is (as you point out is essentially) determined by the drain resistor of the diff pair.

The second stage open loop bandwidth can be supported by increase feedback but only if sufficient overall open loop gain is available which is largely dependant on the open loop gain of the input stage.

Maintaining appropriate open loop bandwidth of the second stage (without much feedback) dictates a low source resistance and this in turn requires a given bias current ie 20 ma approximately for voltage requires across the drain resisters ie 4.3 volts to operate the gates of the output stage

As you point out Patrick, it is intuitive that using parallel input devices supports the required bias current and open loop gain.

Increasing the number of parallel input Jfets would further increase the open loop gain but reduce open loop bandwidth.

I have not attempted 3 pairs of devices because of the complexity but may consider it .

I am interested in your thoughts on the potential for improvement?

I have been running the amp with 2 pairs of 2SJ74 continuously for 24 hours .

Some vital signs.

Voltage across the 392R is 4.55 volts ie 11.66 ma
Relative offset 24 millivolts
Absolute offset 0.14 volts (14 volts at turn on with 68R output to ground resisters 22K McMillian resisters)
Supply rails 23 volts
Total bias = 5.37 amps

You might think 14 volts at turn on is a lot. I guess it a but it drops down quickly within a few minutes to less than 10 volts and over the course of one hour to under 200 mv. (I can see why Neslon used 30 parralell 1000R resisters in the AX200!)

I assume that the bias noted in the working circuit is a reasonable working point for an overall ids of 18.8 ma.
Please comment with recommendation if otherwise.

Ian

[EUVL]

In the standard Aleph X circuit, you have two frequencies to worry about -- one larger determined by the apparent input capacitance of all the Power mosfets in parallel and the drain resistor of the first stage plus the equivalent gate stopper resistances of the parallel power FETs. But the apparent input resistance is also affected by gain of the second stage due to Miller effect, which depends on total power FET transconductance and output impedance.

But apparently, for reasons that I still do not fully understand, Nelson prefers to use large number of devices than necessarily, which (IMHO) does not improve transconductance much when maintaining the same total bias (but just divide that between more devices) as the FETs are still in their parabolic regime, but add capacitance proportionally to no. of devices. So maybe there is a optimal frequency to be had which is not necessarily high.

While you can lower the first stage drain resistor by paralleling more input JFETs, you also increase the input capacitances of the diff pair and hence lower the input stage bandwidth.

How these affects one another in closed loop is best investigated in Spice or another simulation software. I would encourage you to do that first before running more JFETs in parallel.

I personally no longer use the Aleph-X standard circuit but have converted it to an Xed JLH Mosfet as I published in the PLH thread, in which I use a single 2SJ109 as diff pair with 910R drain resistor, a phase splitter with cascoded 2Sk170 (because of voltage) with 500R drain and source resistor and 3x 2SK1529's in parellel as power output MOSFETs. And of course the 2SK1529 has much lower capacitance than the IRFP240 for about the same transconductance at 1.3A bias. So essentially you can still use 1 pair 2SJ74 and 900R drain with Toshiba power FETs and get the same frequency characteristics as 20mA bias and 392R drain with IR devices. My close loop bandwidth (-3dB) is 350kHz into 8 ohm resistive. Toshiba FETs sound different. It is very much a matter of taste. And there is no cheap and easy way other than to build and listen

And I would also encourage you to test (simulate) using a reactive load as proposed by Stereophile. I have published the linke somewhere on some dummy load threads over at the Solid State Forum. Please do a search.


Patrick

PS I have 2V absolute offset when cold, within 10mV between 1 hour and 2 hours, and less than 2mV relative offset. I find your DC offset high, but then I am using 10k for absolute DC feedbacl to diff pair source.

[EUVL]

Perhaps one more comment on MOSFET choices, to at least try to be neutral.

It is noticeable recently that quite a few well-known gurus are turning to Fairchild MOSFETs. Nelson has mentioned Fairchild equivalents to IRFPs in the F5 article. Just the other day, John Curl also stated in the Blowtorch thread that Blowtorch has Fairchild (and not Toshiba as widely speculated) MOSFETs. I believe that Grey is also using Fairchild for his GR-25, if my memory still works.

On the other hand, there are still quite a large number of known designers preferring Toshiba devices. They include ASR (Emitter II uses 2SK1530 / 2SJ201), Borbely, Kaneda, etc. There are also many in the DIY forum who use Japanese devices like Toshiba, Renesas and Sanken. Again, it is a matter of taste.

I have not tried Fairchild as yet, but I prefer Toshiba to IR in the few circuits I built. And I think I am not wrong in saying that their complementary devices, especially those specifically for Audio Applications, are MUCH better complementaries than those from IR.

Of course as many here also know, I also like the LU1014 a lot, which is neither IR nor Toshiba nor Fairchild, is available still only from one source (in Germany), and has no complementary P-device. I have used LU1014 with bias from 200mA to 1.3A. Ask Steenoe. He wouldn't swap his DAO Follower for anything. But then that is another story.




Patrick