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

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.

fab

Member
Joined 2004
Paid Member
...
FAB, paralleling input devices might work, but it would be nice if we could avoid it I think. Do you think that sound wise paralled jets would still be better than a single MOSFET?

It was only a suggestion but I have never tried paralleling jfets neither a mosfet as input. My latest experience with a single jfet as input with about 70% IDss of bias, mosfet VAS and Lateral mosfet in class A (1.1A) gives plenty of bass but I have a big power supply too...

Low source resistor for the jfet as you finally did is the way to go.
 

fab

Member
Joined 2004
Paid Member
I know now what is a Hawksford Cascode. Thanks.
What happens if we have not clean power rails to feed Cascode?
I want to keep Fetzilla somewhat SIMPLE.
I do not like idea to complicate. Just for some small improvement.
But let me try to simulate and see how much improvement with Hawksford Cascode.
And anyone is of course free to test and use it.


Regarding MOSFET input vs. JFET input.
I was surprised to see ZVN3310A had a little less THD than 2SK170BL.
MOSFET ZVN3310A are surely nice small mosfets.
I was sure the JFET had less THD.
But this is not a fact in my simulation.
 
With MOSFET ZVN3310A input.
And AKSA suggested bias of the input.
And BJT BD140 VAS.
Feedback divider 330/22 Ohm.
Output biased 100mA.
 

Attachments

  • test_fetzilla_4c_mos.png
    test_fetzilla_4c_mos.png
    18 KB · Views: 647
Yes lineup,

My simulations show the same thing but I have been nervous to mention it. The MOSFET, at least in simulation, has lower THD then the JFET and similar distortion character. Actually, the FET had half the distortion if I recall correctly (though both were very low). The models may not be accurate though so we need to be cautious.

If they are accurate, whether this translates to better sound I do not know yet, but I am leaning that way. We need to make sure we aren't trying to make a square shaped tool fit a round hole. The mosfet just might be better in this application - at least with the circuit kept simple with no cascodes, etc, and we need to be willing to accept that possibility. But I'll give the JFET a very good go before deciding. I suspect the improved distortion figures are purely due to the much higher transconductance and if that is the case there are probably even better devices to use. The depletion mode device hugh posted had 300mS (3x the zvn3310a and 15x the 2sk170)!

Here is the current circuit I am using with the JFET. Please recall that it was originally the same as the mosfet version in post 336, but had limited bass performance. Dropping the feedback network from 1k/100R to 200R/20R to get more gain in the input stage has improved the bass slightly, but my feedback capacitor is now too small - as Hugh suggested.

I think by making the feedback capacitor even bigger, maybe 2000uF as Hugh suggested, we may be back on par with the mosfet. Currently with 470uF, in simulation we are losing a lot of gain at 50Hz, about 1V on a 15V sinewave. I will try this tomorrow. It's been a big week and I badly need sleep - this is just too much fun :)

With the mosfet version (see post 336) and 1k/100R, 220uF still had good bass because the current through the feedback shunt resistor was much smaller. 220uF was adequate, though still a little small.

Also, decreasig the VAS degeneration to get some more OLG might help bass too, but then we also increase distortion and instability.

Or, as mikelm suggested, we could revert back to a fet VAS to get some more OLG that way.

For a simple circuit - single ended input with no cascode, single ended VAS, lateral P-P output and no compensation, the mosfet version is looking pretty nice though.

Lineup, you are right, I need to be able to A/B test these things a bit more easily than I can now, as I currently have a minimum of 10 minutes between tests.

However, I can say with absolute certainty that the mosfet version does indeed have more bass as certain things in the room shake with it that don't with the jfet version :)
 

Attachments

  • Image 014.jpg
    Image 014.jpg
    97.8 KB · Views: 648
Last edited:
Maybe we should try a BJT input, too.
BC550C might give those FET a good game :D

Yes please do! I'm off to bed but will try models tomorrow.

BC550 = cheap, low noise, plentiful.

Pinout is the same as the ZVN3310A and 2SK170. I can drop it right in if your models suggest it may be worthwhile.

Isn't it interesting how we have gone all over the place with this design - singled ended, LTP, jfets, mosfets, BJTs, cascodes, compensation etc. How educational it has been for me.

When we have a final design I promise to produce a PDF complete with schematic, DC offset plot, stability test plots and FFT, because now we are up to over 400 posts and all the data will be all over the place and difficult to follow. I will borrow some instruments to get a distortion spectrum and THD.

I hope everyone partaking or even just reading along is enjoying the ride. I wish someone else was making one too because I hope my listening tastes aren't taking us in the wrong direction!

At least most of the pinouts are the same, so when we finally make PCBs we can swap parts easily. I know mikelm is going to want to give the FET VAS a go, and others will need to try the JFET input :) :)
 
Last edited:
Swordfishy,
thanks for your elaborate answer, that does indeed shine some light on things!

I myself seem to have a faible for yamaha, both my vinyl rig and my drum kit are made by them and both I admire a lot.

Hugh,
earlier in this thread you mentioned you didn't like cascodes soundwise. I have not much experience with poweramps, in low level high gain phono amps I experimented a lot with casc. and every time I added one atop a 2sk170, 2sj74 or a 2sk369, things improved a lot regarding clarity especially in the lower registers. Cascodes can oscillate nastily, and that screws sound quality. The way you bias them at the bases makes an difference as well, soundwise.
Would you mind to elaborate your findings re cascodes in power amps?


Rüdiger
 
BC550 = cheap, low noise, plentiful.

Pinout is the same as the ZVN3310A and 2SK170. I can drop it right in if your models suggest it may be worthwhile.
Here are simulation THD at 1kHz 1 Watt 8 Ohm.
All other things equal for each transistor
Code:
THD % - Transistor input
0.003170 - BC550C (BJT) Winner!
0.003965 - ZVN3310A (MOSFET)
0.004620 - 2SK170BL (JFET)
Comment:
Differences in THD are so small, that other qualities like Harmonics spectrum and result at other frequencies and/or power levels can be used to decide what input transistor you use.
 
Among the audio elits like Nelson Pass, John Curl, Erno Borbely etc it has been well understood for many years that although jfets consistantly give higher distortion they generally sound better. I think we just have to tweek this design to make it sing.

Personally I think it is a pity we moved away from DC linking, for me it just sounds much better across the whole frequency spectrum, generally cleaner and also better defined, tighter bass - perhaps we are throwing out the baby with the bathwater by playing conventional & safe. ? ? ?

The problems as I understand them of DC linking are:

1) DC offset - TimA, on the JLH simple class A thread has already solved this problem - I'm sure his idea can be adapted to this design.

2) Protecting a Jfet - well, after trying many fairly exotic film i/p caps, my conclusion was that "the only good cap is no cap" - so I'm prepared to take the risk of a "DC event". I'll also lower the input impedance to no more than 10K ohms and possibly a lot less - this should help a bit in this regard.

3) Worry about damaging the loudspeaker. For this I will use a small length of constantan wire soldered between 2 pins as a 0.1 ohm o/p fuse. This wire sounds as good as any o/p resistor I ever tried and much better than most and is very cheap.

If I missed anything that anyone thinks is a deal breaker here - please let me know but for for now at least there is no contest, I'm going for a DC linked version !

After pouring over hundreds of simulations and hearing Swordfishy's feedback, here is the circuit that will be my starting point - apart from the BJT VAS it's almost identical to Lineup's design on page 3. For me, the simplicity elegance of this design is too compelling to ignore.

I am also planning to experiment with combinations of fet VAS, CCS i/p load resister and i/p stage Hawksford cascode and if absolutely necessary a servo.

Oh and by the way, I would not expect the benefits of the Hawksford cascode - if there are any - to be apparent in spice. My use of this would be aiming at achieving constant power across the i/p device at all times so it may have to be used in conjunction with a CCS.

I can already hear some people saying - oh but that is getting so complicated - but for me circuits should be as simple as possible but complicated enough to sound great . . . and there's no harm in trying various things to see how they sound :)
 

Attachments

  • Capture-1.jpg
    Capture-1.jpg
    282.4 KB · Views: 689
Last edited:
MikeLm.. could you possibly link to the JLh simple class A thread where the offset problems have been solved...

On simulation the higher order distortion drops by a magnitude if you use hawksford Cascode..and you also gain som better phasemarigen and additional gain so it should be easier to keep from occilation...In my simulation i have 3 green leds to set the Casodevoltage from the buttom for the Jfet...

Would like the DC coupled version.. but to be honest the lower DC gain by having a cap in series with tha feedback gnd resistor makes DC-triiming quite easy and not so prone to drifiting...
 
mikelm

I am glad you will test the DC-version.
This was my first idea and is surely the most elegant.
Hope it works.

I see you use a special compensation. They say it is better.
You use rather heavy degeneration of VAS transistor.
Might contribute to stability of DC-offset.

C5.
I actually simulated testing 10u 4.7u 2.2u 1uF and no cap in that position.
2.2uF was the winner in low THD at 1kHz
So, 1uF is close. Can probably vary because of the size of R10.

The gain 750/47+1 gives 17
For 1.41Vp input we need x14 to get 20Vp (25 Watt)
680/47+1 gives x15.4
... so this I suggest


I tried simulate with 2 RED LED and 1mA Hawksford cascode.
Did not show any improvement. In fact THD became a bit higher.


If you use one MOSFET for VAS, you can improve by replacing R13 potentiometer
with one 2SK170 coupled as variable current source.
MOSFET has several Volts VGS and so one can fit in JFET there.

See diagram.
25 Watt with low distortion.
 

Attachments

  • test_fetzilla_5e_25w.png
    test_fetzilla_5e_25w.png
    21.3 KB · Views: 665
mikelm.. could you possibly link to the JLh simple class A thread where the offset problems have been solved...

here is a link to the class A amplifier web site

The Class-A Amplifier Site - JLH Class-A Update

Read the text below the second diagram.

here is a small extract . . .

"Tim has been able to achieve an output dc offset variation between switch-on and normal operating temperature of less than 50mV."

This will take a few variations to the current diagram but I think it is definitely do-able.

I won't experiment very much in spice before I build because there will be no drift to assess

mike
 
mikelm

I see you use a special compensation. They say it is better.
You use rather heavy degeneration of VAS transistor.
Might contribute to stability of DC-offset.

the extra degeneration decreases distortion but does not necessarily decrease OLG - it's a complex system but all in all it seems to give benefits in all significant areas

C5.
I actually simulated testing 10u 4.7u 2.2u 1uF and no cap in that position.
2.2uF was the winner in low THD at 1kHz

thx - I'll try 2.2uF also

The gain 750/47+1 gives 17
For 1.41Vp input we need x14 to get 20Vp (25 Watt)
680/47+1 gives x15.4
... so this I suggest

Yes - but DVD's seems to recorded at a lower level so I may even need to go higher

I tried simulate with 2 RED LED and 1mA Hawksford cascode.
Did not show any improvement. In fact THD became a bit higher.

I'll check the Higher order levels but I also found stability was improved but ultimately I'll let my ears be the judge

If you use one MOSFET for VAS, you can improve by replacing R13 potentiometer
with one 2SK170 coupled as variable current source.
MOSFET has several Volts VGS and so one can fit in JFET there.

See diagram.
25 Watt with low distortion.

Thx - I have tried this and I think I could fit in also with a BJT with 200 ohm degen res but I need to check

Also just wanted to say thanks very much for such a great little circuit idea I can't wait to build it.

mike
 
Last edited:
MikeLm.. could you possibly link to the JLh simple class A thread where the offset problems have been solved...

On simulation the higher order distortion drops by a magnitude if you use hawksford Cascode..and you also gain som better phasemarigen and additional gain so it should be easier to keep from occilation...In my simulation i have 3 green leds to set the Casodevoltage from the buttom for the Jfet...

Would like the DC coupled version.. but to be honest the lower DC gain by having a cap in series with tha feedback gnd resistor makes DC-triiming quite easy and not so prone to drifiting...

Actually MiiB is right.

The Hawksford Cascode does lower the distortion as output admitanced goes up because of the lack of BE modulation.

But implemented the wrong way it could make the amp prone to oscillation. Reallife inductance can play with you.
 
perhaps 2 birds with one stone ?

This measure lineup suggested may help a lot with DC offset on a DC linked version and inject a bit more life into the amp with a jfet input.

It adds about 6db of OLG just as Hugh predicted giving about 77db OLG from DC to about 1Khz

The compensation measures are not optional with this design but with the 1K 1nF i/p filter you will hardly see the difference.

The 200 ohm VAS degeneration resistor may also be compulsory but I did not check.

Would be great if swordfishy feels like checking this ;)

An output snubber of 4ohm & 100n allows the compensation cap to be smaller perhaps 200pF rather than 330pF
 

Attachments

  • Capture-1.jpg
    Capture-1.jpg
    283.7 KB · Views: 812
Rudi,

My experiments with cascodes were performed some years back on power amps using LTP input stages and before I began to use LTSpice. I did not notice oscillation at all, but there was no discernible improvement in the sound that I noticed. I concluded that cascodes should be used only where the voltage rating of the input device, such as a jfet, was not adequate for the job, and that the best of them was the Hawksford cascode. For my personal preference for the lowest possible component count, I came to the conclusion that they did nothing of use to the signal path, but this was my experience, and clearly is not yours.

I did use a cascode in a MC headamp I designed, the Paris. But it is a partial Hawksford cascode (modulation with 50% of signal, not 100%), and designed for low noise and a particular distortion profile.

Mike, 200R of degeneration seems a lot. If the VAS runs at 11mA, then re is 26/11 = 2.3R. In my experience, ten times this resistance, near on 22R, is sufficient to swamp the non-linearities of this intrinsic emitter resistance, and further degeneration creates too much current variation in the first stage, detracting from its linearity due to Vbe/Vgs changes on the input device.

For some reason a number of posts were silently removed without trace in the last 12 hours. I do not know why this happened, but I would hope there is a reason for it.

Like the direction of this circuit, it's very good....

Hugh
 
All,

Again, I have told a fib. I think, theoretically, a small feedback cap should improve bass by reducing input stage degeneration at low frequencies and increasing OLG. So my idea posted at the early hours of this morning do not apply.

I have spent all morning experimenting and believe I have cracked this amplifier. I have tried, to name a few things:

Different sizes of feedback capacitors, including NO capacitor at all,
Different currents in the VAS and input stage,
High Degeneraton and low degeneration in both the input and VAS stages.
I have tried mosfets, JFETs and now even a BJT in the input stage.
I have tried different types of feedback networks.
I have dealt with DC offset issues, almost incurable oscillation and all kinds of compensation.

I have dealt with all of these simultaneously with different devices. Now I think I know how to make this amplifier sound good without resorting to further circuitry.

The rule is, use high gain devices then degenerate to get it stable.

The amp appreciates a high impedance feedback network, i am using 1k/100r. I think this due to the fairly high positive voltage required to bias the input with such a feedback network. I don't know why, but the bass performance is better this way. Maybe hugh can explain. This is what I first used with the MOSFET design that I liked so much, and now also with the bjt design I am currently listening to. Unfortunately due to the relatively high degeneration this creates, a jfet is no good here. You need to use a MOSFET or bjt for the input stage. The bc550 verges on clinical, although the bass is excellent. I suggest the MOSFET for a bit of warmth.

I tried to get the jfet to work, I really did, but I think without a cascode or whatever, it just doesn't cut it. I tried reducing the input and vas degeneration to get a bit more olg, but I just ended up with oscillation issues. A miller cap didn't help the sound at all.

Without resorting to additional active devices, I think I have found the best design for this amplifier, and really, it sounds excellent. I will post two schematics soon.
 
Last edited:
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.