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.
This sort of oscillation is difficult to find, SWF. I don't think it's the VAS because increasing the miller cap makes no difference. It's true there is no input filter, but instability is instability and would still be there with such a filter; it must be found and prevented.

Wahab's solution is a good one from a linearity POV but I'm leery about the sound quality. It would be a technically superb version, that is certain, and I think it would be stable but the CCS on the jfet source and drain, the EF driving the bipolar VAS and the TMC are radical departures from the original topology.

If you want to stay with the original topology, I'm thinking a 500R base stopper on Q1, the active element of the CCS for the VAS, would cure the oscillation.

Keep on truckin', this is where it gets interesting.....

Hugh
 
Hi all.
If the problem is the miller cap because they do not use an RC filter at the input of the VAS?
GEirin

I think this would work, but adjustment will be more involving.

1: Set the cap to a large value (say 10uF), then lower the resistor until oscillation stops (start with a 1k resistor).

2: Lower the cap until stability begins to worsen, and then return it to a satisfactory value.

- keantoken
 
SWF, in simulation I've found the positive rail decoupling caps cause a resonance around 4KHz, right in the most sensitive region for the ear. You may want to investigate this and see if the prototype shows the same anomaly. It maybe be easier to detect if you feed a square generator into the output of the amp through a film cap and watch the scope for ringing (this is a great way to detect any anomaly really).

- keantoken
 
Anyway, I'm having a little oscillation problem on the negative half cycle. I'm thinking it might be due to the VAS CCS oscillating. The current circuit is shown below, but I have tried up to 150pF of miller capacitance to no avail. I'd be interested to know what you think might be the cause.

Did you try just going back to "slower" devices ? i.e. irf9610 &/or BD139 ?

Perhaps you did this already but I would try switching back to BD139 as a first step.

I'm going to build with the slower devices - because a 200kHz nice square wave is fast enough for me !

Bichi,

thx for the data sheet etc, I tried two before and neither stated pin out.

mike
 
All,

There have been a few posts I need to respond to, but I am about to head out the door and I really just wanted to mention something, especially to mikelm who is in the process of building. I will respond to the others later. My apologies.

The IRFP9610 while sounding nice, just does not compare to the ZVP3310A. If you can get these (digikey has them), you owe it to yourself to use them.

The difference is absolutely night and day. Like going from 2N3055s to something modern. The ZVP3310A really is unbelievable.

I have been avoiding them because I wrongly believed that I wanted at least 10mA in the VAS, and even this figure is pushing the dissipations limits of the ZVP3310A.

However, my experience is with vertical fets, and I did not fully appreciate just how easy laterals are to drive. I am running a simple diode/bjt CCS in the VAS and the ZVP3310A at maybe 6-7mA. Even at this current it can do a reasonable 20kHz sq wave, but more importantly it sounds absolutely amazing. It makes a world of difference, really.

Another problem with this device, and partly why I have avoided it except for a little while at the outset, is that it makes a damn fine oscillator.

I am currently running it with no compensation in a quasi stable state (stable only when a load is attached). It is the best amplifier I have ever heard.

Also mikelm, I urge you to try whatever you can to avoid compensation, especially the miller type. It has an audible detrimental effect and seems to remove some of the punch from the music.

Try a little more CLG or less OLG before you resort to compensation. I think it's worth it.

Mikelm, having led you down the IRFP9610 path, I would be happy to post a few ZVPs to you to try if you like.

Anyway, the other half is getting impatient. Will post more tonight and respond to everyone.
 
Last edited:
Keantoken, Hugh,

Thanks for your replies.

I actually got the amp a little more stable by removing the compensation altogether!

Anyway, need to run, but yes keantoken you are right, I forgot to update the schematic and still had the 20mA figure there from when I was running a different CCS. I am now running about 6-8mA with a LED/BJT ccs. Same kind of oscillation as before (but only with no load attached).

Keantoken, your post says feed a sig gen into the output? Am I reading this right? Do you really mean output or input?

OK, I'm about to get strangled. Gotta run.
 
Last thing!

I'm thinking that given that it is stable with a load attached, an output snubber (which I currently am not using), might fix things. However, it will have to be pretty solid as I almost set fire to a few 2W snubber resistors last time I was fiddling with the ZVP3310A! Seems a bit excessive.
 

Attachments

  • UGS3.jpg
    UGS3.jpg
    75.5 KB · Views: 404
  • PassLabs UGS 1- to-4_Pagina_10.jpg
    PassLabs UGS 1- to-4_Pagina_10.jpg
    111 KB · Views: 391
An output snubber seems right to me too. Maybe you should try carbon comp resistors here?

Another option is to add a snubber to the VAS output. Actually, isn't that effectively what you are doing when you downgrade the CCS? Is it possible that you could keep the good CCS and just insert an equivalent snubber? This is looking more and more to be the solution to me.

Have you thought about using a ground plane?

You are right, feed the signal generator through a large film cap (to block DC) into the output of the amp. How sensitive is your scope?

I discovered that cascoding the Jfet helped the 4KHz bump by about 30db. Apparently this is an issue with PSRR. Removing the rail caps completely eliminates this issue, but obviously that is not the solution we want. One possibility is to power the frontend on different rails. These regs could be set up quickly:

http://www.diyaudio.com/forums/powe...ntokens-cfp-cap-multiplier-2.html#post2371837

I think to get rid of the 4KHz bump cascoding or regulating may be the only option. Cascoding the VAS makes it worse.

Incidentally, removing the filter caps entirely improves the harmonic profile, by increasing even harmonics. I think you should add .5R series resistance to your supplies in simulation like I did, because this changed some things. Is it possible that optimistic simulation is showing an optimistic profile?

- keantoken
 
It seems that there is positive feedback between the output MOSFETs and the Early effect of the Jfet. The inductance characteristic of the amplifier's output impedance may be resonating with something in the amp that is producing a capacitance multiplier effect. Or it could be something stupidly simple that I never thought of. However when the input+VAS and output stage rails are separated, the effect seems to go away.

- keantoken
 
I just tried 100R+1mF filters on both rails to the Jfet+VAS and it didn't work. But I just figured it out.

There is no resonance, just a convergence of three curves. The PSRR of the amp, the rail AC response, and the output impedance of the amp. Reducing the rail filtering to 39uF eliminates the dip. Alternately, increasing the filtering by a great amount will increase the amp's OLG and decrease it's output impedance by a large margin.

There is no resonance, but what I have discovered is that there is a large gain to be had from regulating the input stage rails. Distortion and output impedance decrease by a large margin. Furthermore, if the amp is left with such low PSRR, the filtering caps' characteristics will have a large influence over the final outcome.

- keantoken
 
Wahab, please verify your output MOSFET models, I seem to remember those models were flawed. I think TMC should be explored only after the circuit is stable

Well , i think that the models are quite accurate enough , but who knows..
Would like to try with the models you re using since you ve got also
the Hitachi s ones.

Anyway, using any hexfets from IR will yield exactly the same results.

As for stability, it is implemented thanks to this very same TMC network
that allow robust compensation while preserving the potential linearity.

Indeed , this kind of topology is often dismissed due to its inferior
linearity, but with enhanced VAS and TMC , it can be classified
as high definition audio amp with THD residual below 100db ,
surpassing the CD noise/thd usual ratios , as should any amp
if it is to be used with such sources.
 

Attachments

  • FETZILLA MODDED SCH.gif
    FETZILLA MODDED SCH.gif
    8.9 KB · Views: 492
  • FETZILLA MODDED HEXFET OS SCH.gif
    FETZILLA MODDED HEXFET OS SCH.gif
    9.2 KB · Views: 485
  • LFET VS HEXFET THD10.gif
    LFET VS HEXFET THD10.gif
    36.5 KB · Views: 477
Wahab, add .5R series resistance to those 25V supplies and add however much filtering you like, and look at the specs again.

- keantoken

With 0.5R and 4700uF , the THD is exactly the same , as can be expected
due to very high OLG and the resultant PSRR that increase as well with
available loop gain.

Or are these RC circuits there to test other parameters ?

Not sure about what is to be measured.
 

Attachments

  • FETZILLA MODDED HEXFET RC SCH.gif
    FETZILLA MODDED HEXFET RC SCH.gif
    9.8 KB · Views: 456
  • FETZILLA MODDED HEXFET RC THD10.gif
    FETZILLA MODDED HEXFET RC THD10.gif
    36.8 KB · Views: 444
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.