Jfet/bipolar PCB and Schematic (finaly)

First the schematic...

Match Q1 and Q2 (J111 - J113) to a current difference of 100uA when measuring IDSS.

Change R1 and R2 to 470R.
Change R3, R4, R14 and R15 to ~ 200R
Change R10 and R13 to 500R (multiturn)
Change R18 to 2K0

Set ID on Q3 to 6 - 8 mA.

Set ID on Q6 to ~ 6 mA.

Set ID on Q2 o ~ 2 mA.

Then it should idle around 30 - 35 mA.

The feedback is outside the PCB. Remember short tracks!.

Slewrate is ~+/- 75V/usec.

at unitygain -3dB is ~4MHz.

It is stable with 1nF load (No overshoot! at squarewave signal).

Try it!, i have mounted it inside my CD player and it have in my system better treble perfomance than OP275GP when driving
1K || 220pF load.



  • scm.zip
    77.2 KB · Views: 8,961
I forgot to say :

You need to glue the following together to minimize drift:

Q1 and Q2
Q4 and Q5
Q7 and Q8
Q11 and Q12
Q13 and Q14
Q15 and Q16

And when you match Q1 and Q2 to around 100uA you will get an offset on VGS around 10 - 20mV.

Buy 100 J113(It will cost you about 500 DKR.) and you will get lets say 20 pairs! It is cheaper then buying U401,U404 etc....

And the noise level is typical around 3nV/Squareroot(HZ) @ 1KHz.
Not as good as 2SK170 but cheaper!

The Last part!

Here comes the PCB and PCB placement on top and bottom (SMD resistors in 1206 housing).

Happy DIY to your all!

Is anyone interrested if iwas going to make it smaller? Lets say 2 - 3 times (in both direction) the size of a DIP8 house used by the most opamp with 8 pins, so it could be mounted instead of an OPAMP.
The it would be made with SMD bipolar and maybe SMD JFET's.
It will be a doublesided PCB.



  • pcb.zip
    78.2 KB · Views: 3,236
Nice Circuit!

Hi Sonnya,
Congratulations with your circuit. It is the first discrete <I>folded cascode</I> design I see on the net!
I have a few suggestion and questions:
I think it is better for the <B>sound </B> to use a monolitic dual JFET like 2SK389 or industrystandard 2N5912. Any reason why you use FET's as current sources? I my own design I use LED/transistor current sources. Personally I would raise R3 & R4 more to say 2k and lower R1 & R2 to 10 Ohm but that's only a minor detail. The matching transistor issue also applies to the current mirror Q7 and 8. Would use a dual here too.
Are you familiar with the Application Note High speed Amplifier Techniques? It contains the the same type buffer you use and some very practical tips to prevent oscillations

You are right about the Fet's. 2sk389 is a good choice .. lower ciss and crss. Lower noise (1/3 of J113). better cmrr and higher drain voltage (J113 : VD = 35V).

But the 2n5912 is easy to get here in denmark.
Same noiselevel as J113.
Matched to 15mV.
Drawback!? drainvoltage of 25Volt.

I think i will try 2n5912 in the next weeks.

Why i use fet's as currentsource?
1. to test how stable they are. And the J113 has a good temperature stability at current level around 2 - 8 mA.
2. the have a high gos (output admitans) when the have a Rs in commonsource configuration. Higher than bipolar types.

rd = 1/gos.
Ro = rd*(1+gm*rs).

The low voltage noise of fet's will also show of in the current noise when connected as a current source.

But please correct me if i am wrong.

The thing about r3 and r4. If you raise them to 2k then the current for Q1||Q4 and Q2||Q5 is as low as ~ 550uA. If we do this we have to lower the current generator Q3 to ~ 550uA.
Then the current flow through Q1,Q2,Q4,Q5 is 275uA each.

I have tried started out with r1 and r2 set to 2k each. It gets slower, and the open loop gain gets lower also. So 10R0 should make it faster!. But i haven't tried it out.

mmmhhh Q7 and Q8 i have also thought of using a mathed pair like dual's but then i am limited to only a few ones. But your are right about matching them.

Other types then 2n5551 and 2n5401 will maybe have a higher output admitans. Something to try out!

The reason why i only have matched the J113 Q1 and Q2 is to see how it behaves before i go to sot23 housing(Try match smd transistors ... yerk!!!)

Thank's for the app.note. I have made some study about what analog devices says about highspeed design's. But Linear does have some know how to!!!

Thanks a lot for the feedback!
JFET input/folded cascode design

Hi Sonnya,
<B><I>InputFETs:</I></B> You can also use the 2N3958 with 50V breakdown voltage. It is in a small TO-71 housing. But I wouldn't be too concerned on the 25V breakdown voltage of the 2N5912 as the current source takes a large part of the voltage drop.
<I><B>JFET's as current sources:</I></B> May I quote from Horowitz and Hill, the Art of Electronics , Chapter3 $3.06 about JFET current sources? "It is important to realize that a good bipolar transistor current source will give far better predictability and stability than a JFET current source" See also the link to PeuFeu France for cascoding current sources.

<B><I>Analysing the currents through R3 & R4</I></B> As I see it the curent through the resistor R3 is the sum of the currents through Q1 & Q4. The current through Q1 is determined by current source Q3 and is just half of it. The current through Q4 is determined by the voltage drop of the LED and the rest of the "divider" LED/R11, Q6. In fact Q4 and Q5 are a current source in itself! I would omit Q6 making the voltage at the base of Q4 & Q5 adjustable by means of a adjustable resistor (pot) in series with the <B>LED</B> or use a string of LED's. To give a example. If the current source Q3 provides 2mA Q1 passes 1mA. And if the current through Q4 is 0.5mA. the current through R3 is 1+0.5=1.5mA giving a voltage drop of 1.5mx100=150mV below the supply voltage. Also note that the voltage at the bases of Q4 & Q5 should be 0.65V lower than the voltage of there emittors, point 2 at R3 & R4.
You see the example I gave is not correct as the voltage calculated by substraction of 0.65V of the emittor voltage will give 15-0.15-0.65=14.2V. While the LED will give rise to 15-1.6=13.4V !. I assumed =/-15V supply's

Interestingly most of the gain of the circuit comes from the current mirror Q7 & Q8............

It must have been a revelation for you to that a simple circuit like this is sounding much better than a highly praised opamp OP275!:)
I have read you thread. And i will get back later today. with some answers.

First : Yes! Supply voltage is +/-15V that is right and i have forgotten to mention that!.

About the gain in the circuit. It depends a lot on the Q5 and Q8 output admittance. Does Q7 and Q4 only do a current "transfer" to Q8.
The gain is also set by the two resistor R1 and R2. The smaller they are the bigger the current difference will be at a given voltage diffence at the gate of Q1 and Q2.

So the dc-gain is nearly a product of : gm in Q1 and Q2, R1 and R2 (is the source resistance on the fet's) and the output admittance of Q5 and Q8.

Q5 is coupled as a commonbase configuration and the output admittance will then be : hob = hoe/(1+hfe).
example :

hfe = 100;
hoe = 10uS; (100KOhm)
hob = 99nS => 10.1MOhm.

I can't remember how to calculate the output admittance of a currentmirror right know .. So it have to wait.

It is right that Q4 and Q5 is a current generator in it self, but the configuration with the LED makes up a constant voltage across R3 and R4. So there will "always" be ~ 1.2Volt across R3 and R4 wich in return will generate a current through R3 and R4 (200 ohm) of ~6mA in each resistor.
So if Q1 sinks 4mA through R3 then Q4 can only sink 6 - 4 = ~ 2mA.

So if Q7 was a resistor the voltage across it would be 180 degress out of phase with the inputvoltage on Q1. Like it would i a normal cascode stage.

By the way Thanks for the links. I will get back to current source thing! There is something to change at this point.

By the way : Parts like ztx651 from zetex should have noise figures as small as 1nV/Squareroot(Hz)!!! Check there appl.note DN11.

Sonny:) :) :)
Folded cascode is an old hat

When I had a look at on old cheap Luxman integrated amplifier from 1984, called L410, I realized that it had cascoded jfets int he input stage, and also a folded cascode in the voltage amplification stage, of the discrete power amp section.

And that folded cascode was designed quite good: they use current sources for biasing instead of resistors R3, R4 in the circuit presented in this thread.

I tried a simple folded cascode some years ago, when I had a lot of talk with Jonathan Carr (designer of Connoisseur preamp and Lyra cartridges). I tried it in phono and line stage of a preamp. While treble was very good indeed, stability in the bass and soundstage was too vague for my taste. I assume there is a lot of thermic distortion when biasing folded cascodes with resistors only. Any voltage drift will transform into a current drift, and then the poor input transistors have to rule that out, you see ...

Jonathan Carr uses current sources, too.

Hartmut from Munich
Yep and they have been used in opamp for ~ 10 - 20 years maybe more!
But i like to test different circuits!, and thats why i made this one! :)

" I assume there is a lot of thermic distortion when biasing folded cascodes with resistors only" ... You are right about it ... Resistors change value 50ppm/degress celcius (typical). So big current change -> big power change -> big value change!

I like the last link Elso send about "memory effect" they point out that high openloop gain may be the solution to some of the problem. The higher gain => lower current change => lower change in power discipation => less thermal distortion!!!

I am going to like this!! :)

I find this forum good, cause you get new input!

Nice to hear others opinion! ;)

Not a old hat, never throw away a old one before you have a new onr

Hi Hartmut,
Please post the schematic of the Luxman and I will comment on that. I don't agree with you about the "thermic" distortion. If you use monilitic duals for the input FET's and the current mirror you have absolutely no drift and can omit all coupling caps at the input and output, leaving another layer of grunge!:mad:
Hi Elso

It is funny!!!, i did fall over his page i think 3 - 4 days ago.. haha:)

mmhhh i like the 3 - transistor currentmirror he is using ... it have less current difference (if there is any ) than a 2 transistor i am using.

The thing about currentsource.. what should i say .. After som measurement .. Yes you are right about using bjt as currentsources .. They are faster and more stable .. They get there desired current faster.

And the difference from device to device is smaller than with JFET's

I took 200 pcs. more of J113 and tested the Idss .. Min : ~ 8mA - ~21mA it is far to big tolerance from one batch!!!!! :(

I think i will trymount a 2 transistor currentsource and also a zenerbiased transistorcurrentsource..

2.7V zener have nearly the same temp. coefficient (~-2mV/degress) as a transistor.

By the way what have you been up to ... you did mention some thing about your own design .. If want to tell ;)

- Sonny
Current Sources

Hi Sonnya,
In the MAT03 datasheet fig. 14a & 15 you will find current sources, or sinks if you want, of a bipolar transistor and a LED. These have less noise than the 2.7V zener and have about zero tepmco. Both the NPN type and PNP type are illustrated. You can cascode these with a extra transistor like PeuFeu to eliminate thermal distortion or the Early effect or whatever effect.
Laymen are always very surprised and amused seeing those red lights glowing on the PCB when they are peeping through a ventilation hole.

In the MAT02 datasheet a two transistor current source is pictured. I am not so fond of these but these are used by Metaxas Audio in Australia.

Good point about using LED's (GaAs types!? I am going to find out why GaAs!?).
The two transistor currentsource is not as stable as the LED-Transistor solution, but the noise is as low as the one generated from the transistors and the surrounding resistors.

Passlabs have done a smart thing to lower the noise from the LED. Check : http://www.passlabs.com/pdf/aleph/aonoserv0.pdf

and : http://www.analog.com/library/techArticles/dataConverters/pdf/Publication_V-Ref.pdf

The last one is written by Walt Jung and on page 2.18 the make there low noise circuits the same way!

By the way : half the production of MAT03 is Obsolete. it is the same with the SSMXXXX line.

Sidetrack : You have to check the new AD8610AR .. The datasheet speaks for itself (How many opamp's can drive a 2uF load!!!??? Yes 2uF and not 2nF!!!)

Another Link and HiFi Varia

Hi Sonnya,
Thanks for the links; two were new to me!
Please find also below a link to a article by Walt Jung. He uses a similar buffer you do but with LED's in the current sources!
GaAsP is just chemical shorthand for the chemical elements of gallium, arsenic and phosphorus used in the semiconductor. It looks very scientific but it is just a normal red LED [I am a chemist BTW]
I was familiar with the Passlabs article, but not sure what you mean by <I>"lowering the noise of the LED's"</I>. Should I read <B>FET's</B>? Makes more sense to me. Paralleling active devices like bipolar transistors or FET's is a old trick from the conjuring-box of the solidstate engineer-magician. But you can also look at a large geometryFET like the 2SK389 or 2SK 147 as several small FET's paralleled.
I also was familiar with the low pass filtering of the reference noise, though from a other source (OP275 datasheet).
I do not understand: <I>"half the production of MAT03 is obsolete'</I>? Are they no longer manufactured? The EH and FH types are still available! I do know the MAT02 & MAT 03 are better than the SSM2210 and SSM 2220 although the datasheets claim they are the same. They are not, the MAT's have much higher Beta! As found out MarkLevinson (the Company). They changed to the MAT type's with much ADO.
I am sorry but since I work with discrete circuits I am hardly interested anymore in IC's. Why would you want to drive 2µF?. I have a long cable to the poweramp but that's only a few 100 pF.:)
Sorry about that obsolete thing .. It is only the military types you can't get any longer or for a limited time...

And about the passlink: Sorry i have just looked at it again and i am wrong.

But you can make the noisereduction on a LED like it is done with a Voltagereference. By making a lowpasfilter before entering basis on a transistor (1K + 100uF like fig.2.17 in the AD note).

mmmhhh you are a real discrete freak;) it is interresting to see that it is possible to make "opamp" stage wich is stable with large phaseshift at the output.