JFET input, MOSFET VAS, LATERAL output = Perfect!!

[sng001]

Stanley,

Because your fb resistor is a high 10K, necessary to reduce loading on the VAS (IRF9610), any current not passed by the lower left CCS (U5) will pass through this resistor. In fact, only 2mA through the 10K resistor (50% of the drain current through the jfet), will generate the +20V offset you are seeing.

Solution: The CCS is drawing too much current. For the jfet biasing shown, it should draw around 4.1mA. This corresponds to 330R between emitter and negative rail of the CCS, so simply replace the 100R trimpot with a 200R trimpot, OR, increase R15 to 270R and use the existing 100R trimmer.

Cheers,

Hugh

Thanks Hugh,

First at all, I went home & replaced Q1 with a 2SA1145 & it made no difference to the (HUGE) output offset.

Then I read your post & you were spot on. I used a 2K trimpot for R15+R16, I could decrease the offset by increasing the resistance. The final value was 1016R for output offset of around 20mV.

First power on - -330mV
After 1 min : -70mV
After 2 min : -34mV
After 3 min : -7mV
After 4 min : +9mV
After 5+ min : +20mV

To me that is acceptable.

However, when I power off, the output offset goes up to +2.3V, so we will need some sort of speaker protection.

I connect the Fetzilla pre-amp to the source & it amplifies.

Thanks again, Stanley

 

[GregH2]

Tip on the IRF240/9240: Don't run over 38W dissipation; they really don't appreciate going over 40W unless you have forced air cooling.....

Hugh

Yeah, this isn't for a permanent setup, just a but of fun to hear how it sounds

 

[MiiB]

in simulation I see that the version without the CCs for the front end is dependent on a bias voltage. without it changes character with the input amplitude...and that can't be good...so if one is aiming for the total DC-coupled Jfet-version..there must be a CCS for the front-end..I also believe strongly in locking down the work conditions for the devices.. So a CCS is in my scheme of things...without it the higher-order harmonics develops strongly with signal amplitude...(in building speakers dynamic linearity is one of my main focuses)

 

[mikelm]

will your i/p ccs be above or below the jfet ?

i.e. above the drain or below the source ?

 

[MiiB]

The one giving the best SIM results is below....best results in terms of keeping the distortion profile intact under larger signal amplitudes....I belive this is at key to get the liguide like sound you get from the very best amplifiers on the market...
In sims I have a Red LED biased BJT CCS with app 300 ohm under giving a current of app 4 mA...

michael

 

[mikelm]

Some of us where concerned that this would allow neg rial PSU HF / RF noise to enter into the i/p & feedback circuits - thus polluting the sound - so if you find better performance with this CCS perhaps it would be good to use an RC filter to take the neg rail to about -10V. This should be enough voltage to allow your CCS to work very well and will filter the rail noise significantly.

what do you think of this idea ?

mike

of course this would not be necessary if the ccs is perfect, but this is never the case

 

[MiiB]

I think it's brilliant....I do the same in all my power amps...crude rough supply with ample capacitance for the output, and then a voltage doubler, followed by a number of RC-cells to take the voltage down to app 10 V above the output supply for the Class A stages..... here a further R-C cell or two will fit nicely in...and keep PSSRR very high...

 

[mikelm]

I checked this out:

At 22V o/p a few volts below clipping I see no difference

At about 10V I see about 3dB less higher order HD's

Is that what you were talking about ?

There is a far more dramatic difference just biasing the o/p stage into 1 Amp class A which is where I did my testing.

can you post your circuit to clarify this

thx

mike

 

[MiiB]

Iam not home so i cant post the circuit....on my other computer....

I use BF862 and Hawksford cascode...and the trend is clearly evident...The CCs aids at higher amplitudes and higher frequencies as well...

 

[AKSA]

This is interesting guys, thanks for doing the legwork.

I too will look into the sims, see what transpires.

However, this amp is already very, very good. Its high orders are extremely low, and more than 96% of THD is H2, H3 and H4 - and this with only 500mA of Iq. At +20dBU, using the AC coupled variant without lower CCS on the first stage, I get around 0.0022% at 1KHz, which is up there with the best.

My own preference is to use the AC coupled version so as to avoid a servo; I'm not so finicky about caps, either at the input, or at the feedback shunt, particularly if they have a dollop of DC volts across them.

However, I do agree that Lineup's use of a lower CCS (suitably RC decoupled from the rail) is the only effective way to implement DC coupling, with more than 90% of the 1st stage current passing through the feedback shunt resistor. You can manipulate the bias at the gate of the jfet by altering this resistor, of course, but this is fiddling around the edges; for the gross adjustment, you need that CCS.

Hey, this is shaping up very well.... I have a nice pcb ready, one further refinement to put in, then I will publish it. However, I have NOT done the DC/servo version, only the AC version, which Greg tells me sounds a gazillion dollars.

Cheers,

Hugh

 

[mikelm]

problem . . . . what problem ?

excuse me for being a bit thick here - but I really don't get this problem.

what is he problem with not having a ccs from source to neg ?

MiiB says it is necessary . . . but he is using a hawksford cascode so when I tried his idea I added this ccs to sink the current added from the pos rail for the cascode . . . .

which left precisely the same amount of current flowing through the gain / shunt resistor as I had without the cascode.

so if I don't have a cascode why do I need a ccs ?

I was simply planning to choose a couple of jfets that gave me the right current - aiming at about 5mA

seems pretty straight forward to me - so what is the problem ?

Do we think we need more current ? is that the problem ? so we need to sink some extra to bump it up a bit ? won't that make the jfet a bit too hot ? I thought 0.1watts was about ok.

anyway - I look forward to being enlightened when I wake up in the morning

over & out

mike

 

[AKSA]

Mike,

There is no problem!!

There are two versions of this amp: DC coupled, with no fb shunt cap and shunt resistor straight to ground, and AC coupled, with a 1000uF fb shunt cap.

The former requires a CCS from negative rail to assist with offset adjustment. You can then use any 2SK170-BL you having lying around. This was used in Lineup's circuit, though without DC coupling.

You choose whichever one suits. It certainly isn't a problem to choose any one of the three variants here available.

Hugh

 

[mikelm]

Thank you Hugh,

Happy to wake to no problems,

Wonder if Miib agrees,

my guess is not,

but don't know why.



( does this qualify as a haiku poem ? )

 

[AKSA]

No, Mike, you are using the wrong brush.....

Hugh

 

[mikelm]

hahahaha

 

[MiiB]

I just think that Hugh and I aim at different targets and thus have different tools...I'am not nearly as skilled and experienced as Hugh.. and this here is for me my curiosity driving me...I must also admit that i Aim for a more powerful amplifier working with higher rails 60V...and maybe what I See is also a function of that...

I do however in my simulations see that when the amplifier is pushed..higher order distortions develops...and by inserting a cascode and a CCS.. this development is reduced and the higher order is kept low..!!
I do think it's important for an amplifier not to change character with amplitude...hence my desire for the CCS and the modulated cascode.

Higher gain and super low noise in the input is also desirable...so I'am considering at quartet of BF862..

 

[mikelm]

mmmm - I looked at this again in spice again with actual components rather than theoretical components and changing the feedback resitors

from 25 / 375

to 100 / 1500

so that the CCS was more dominant and with a BJT hawksford cascode with o/p biased to 1A

and . . . I did indeed see the higher harmonic noise floor up to 20khz drop from about -150dB to about -180dB with a 10V peak o/p

So Michael is onto something here

but I'm not sure if I could hear the difference between -150 & -180dB

but it might be worth checking it out.

 

[mikelm]

Michael,

I think we just have to be sure that this arrangement is not introducing other influences that we are not measuring in this spice analysis.

I tried it with both a resistor & a pure theoretical voltage biasing the cascode. The pure voltage gave a slightly better Fourier result but are you sure that those LEDs won't add some noise which will spoil the benefits in other areas ?

In the end I think I will have to trust my ears, so I will definitely build the simple version first and see which sounds best.

mike

 

[MiiB]

Often the LED based cascode is used in MC-riaa circuits...so my hunch is that they will not add to much noise...they are quiet as similar resistors...but firmer in in the voltage.

Looking forward to hear your judgement...think this is a great circuit to explore..

 

[AKSA]

Unfortunately that is the reason, regardless of the variability of human perception, that you must build and listen..... it's a PITA, because listening tests are so unreliable, but I generally find that after a couple of weeks of listening to an amp I can figure out what is good and what is not so good....

We must support SWF in his sonic journey at any cost.....

Hugh