j-fet question from the inexperienced

[CBS240]

Hi all!

OK, I have not made an amp with a j-fet input stage. As a matter of fact, I haven't used j-fets very much at all except for some lab experiments in school. I understand how they funtion and I am making a cascoded diff. I have matched the devices to within a couple of percent. Is it better to use the j-fets with a lower Vgs for the differential pair or a higher Vgs, as this affects the Idss value? Also does the cascode pair need to match that closely with the diff pair? Or does it even matter? The amount of Ids may vary on whether Vgs is higher or lower in order to maitain the correct operating point.


Thanx much

[aparatusonitus]

Hi CBS240,

You will find some answers here:

http://www.borbelyaudio.com/adobe/ae599bor.pdf

http://www.borbelyaudio.com/adobe/ae699bor.pdf

Best, a

[jcx]

vishay/siliconix fet app notes:

http://www.vishay.com/fets-small-signal/ssfanp/

an103 constant Current source app note is relevent to sizing fets in cascode operation

basically the upper fet has to have a much higher operating Vgs to give the lower fet enough Vds to operate in the "pentode" region, with perfectly matched fets you waste the gain of the lower fet

the better approach is to use different fets in each location

[CBS240]

Thanks aparatusonitus and jcx!

I have seen that paper by Erno Borbely before but I only skimmed it and didn't study it much as it never applied to the current project at the time.


I put to use the cascade jfet CCS, interesting concept. Jfet's really can be quite useful in audio. So I built a circuit that seems to work quite nicely. Haven't done any tests or anything....not completely assembled, but so far so good. Thought I would have more difficulty in implementation but it seems to be working correctly. All transistors are SOT-23, except outputs, drivers, and Vbe mult. Resistors are 603 SMD except outputs Re and pots. As of now input stage is biased around 35uA per leg. The jfet I'm using is:http://www.fairchildsemi.com/ds/KS/KSK595H.pdf
Curious if these are any good...they seem to work as expected. They are intended to handle tiny currents, relatively, and are quite linear.

The output stage is driven by +/- 18V. The goal here is 25-30W to 4Ohms, but reliability and quality is a must. This can be done with 1 pair of TO-220 output devices. Consequently, there is a 'discontinuity' when the amp clips, as the VAS will saturate or turn off. To combat this, I used a pair of jfets(below input jfets) to limit the current in each side to 45uA so the other side is forced not to cutoff. Kind of like a cross quad, so as to aid in recovery from clipping by limiting the dynamic range of the differential.

The front end is +/- 26V, fed by a voltage doubler + discrete regulator circuit(already constructed) with excellent load rejection. It is essentially an amp in it's self. I have found that clean DC is less likely to unstabilize a circuit than one with PS noise and generally has a better PSRR. Besides, this way I only need one 12-0-12 2A transformer and I can drive the outputs to the rail and still have voltage headroom in the front end to play with.

[CBS240]

Here is one 'experimental' PCB for the front end... I think the worst part of all this is having to match these tiny transistors. I spent the better part of a day measuring and finding matches. Oh well, I can cram a lot of circuit into a little PCB. Homemade "IC's"? Be better when I actually make a PCB for this stuff.

[darkfenriz]

Originally posted by CBS240

The output stage is driven by +/- 18V. The goal here is 25-30W to 4Ohms, but reliability and quality is a must. This can be done with 1 pair of TO-220 output devices. Consequently, there is a 'discontinuity' when the amp clips, as the VAS will saturate or turn off. To combat this, I used a pair of jfets(below input jfets) to limit the current in each side to 45uA so the other side is forced not to cutoff. Kind of like a cross quad, so as to aid in recovery from clipping by limiting the dynamic range of the differential.

Sorry, but the way you drawn it is not going to work at all in my view, because input jfets are forced to have constant Id. If you want clean clipping, there is a better way that I use with just two diodes.

[CBS240]

Hi darkfenriz,

The diodes is an excellent idea I may add these. When the output from the VAS is greater than the lower rail voltage, the VAS would still try to drive the output load and saturate/cutoff. I'm much more concerned that the LTP doesn't saturate/cutoff. The idea behind the current limiting in the LTP is that the limit is 45uA for each leg, but the bias is only 35uA for each leg. This gives a possible change in each leg current of +/- 10uA, which should be significantly larger than the current change required when not clipping, but far from pinch-off or saturation. Niether current limiting jfet is at it's limit so they are both saturated on during non-clipping conditions. The devices were matched and have a Vgs of -0.21V using a source resistor 4K7. The resistor values in the circuit depend on the specificly measured jfet in order to achieve the right currents.
I shall now experiment with adding the output stage and hope that 2 more PN juntions doesn't create too large of a phase difference and become a funky function generator. lol


PS: I hate breadboards!!!....espeacially when they have been used and have shorts inside.

most frustrating sometimes

[CBS240]

I have now built the outputs on a breadboard, a few simple mods, fiddling with the grounding issues there is now not a trace of PS noise in the speaker. The only thing I here now is a tiny bit of white sssss in the tweeters, but you have to put your ear right against it to hear it. I have heard more noise out of name brand circuits. I'm actually quite taken here.

The only problem I seem to have now is a little bit of DC offset to the tune of about -60mV. Is this a reasonable amout? If not, I am kind of at a loss as to correcting it without degrading the signal. Any ideas are most welcome, I will post the entire scematic shortly...

[CBS240]

This is by a great margin the best sounding and most stable amp I have built yet, and it’s still on a breadboard. Hope the picture is legable. It seems like a lot of circuit to get just 25W. However, most of the components cost just pennies, and some only a fraction of a cent. I was stretching the limit with the regulator circuit and used 470uf caps for the voltage doubler but the scope still shows no PS ripple noise from the amp output with 3V of ripple and a DC variation of about 4V with audio playing…cost and space saver. The output of the regulators has no variation. Around 15mA is the current limit, at least that I tested. More possible with bigger power supply, but the these devices are only 3 X 1.3 X 1 mm.

The transistors I use are:

BC850-- http://www.fairchildsemi.com/ds/BC/BC850.pdf
BC860-- http://www.fairchildsemi.com/ds/BC/BC860.pdf
KST5089-- http://www.fairchildsemi.com/ds/KS/KST5089.pdf
KST5087--- http://www.fairchildsemi.com/ds/KS/KST5087.pdf
KST595--- http://www.fairchildsemi.com/ds/KS/KSK595H.pdf

I wrote the Vgs of each KSK595 in the circuit biased at 50uA, 5V. Most were within -0.28Vand -0.19V. There were a few outliers, I used them for CCS. I probably went through about 25 or 30 of these transistors before I found enough matches. The 5089 has a measured hfe of 400! This helps buffer the jfet source follower to drive the emitters of the second differential because these jfets use such small currents, but means there is more variation in hfe of these type so it takes more time to find matches.

I think I may try increasing the feedback impedance in order to move down the low frequency roll-off if it will stay stable. I found that it would not stabilize without some pole compensation. Don’t laugh, but I started with 0.5pf and had to put 3 of these(they’re 402 SMD) in parallel before it would not oscillate. The impendence is high so with 1.5pf it is stable. The oscillation without the cap is around 5MHz. 1.5pf is like 21K at that freq.

This amp circuit topology seems to be much more promising than the others I have built.