An illustrated guide to building an F5

[tpark]

I posted to the main F5 thread regarding a power supply problem and never received an answer. I have a few wrinkles in my design which I will try to describe. I have quite a few old SK147 and J72 Toshiba fets, the good ones, which bias beautifully at 5ma with 10 ohm source resisters. They are limited to 15volts so I cascode them using another similar fet biased with an LM329DZ which is rated at 6.9 volts. I note Nelson recommended cascoding if necessary. This did not seem like a problem to me. I wanted to use the power supply from Zen3 ( the one with the zener regulator) which I have used before with no problems as I have 24 volt toroids putting out around 35 volts. The output side is stock. Finally my problem is this: using the regulator my positive side pins to only 4 volts and no current flows and the negative is correct at around 24 volts and a small current flows. The cascodes are not biased properly on the positive side clearly and the bias is not correct. If I connect the power supply directly to the B+ and B- and by-pass the regulator everything is fine and I can set up the bias without difficulty and no offset issues. For completeness I do not use a printed circuit and use the frogy style with my fets and resistors glued to a ground plane copper board. I presume there is some sort of oscillation going on and my scope is limited in range so I cannot see it. Has anyone else seen this or can someone point me to a forum where this has been discussed.
Tom

[cviller]

Try to post a schematics of you modded setup (AMP and PSU) and write in the voltages you have measured. Based on your describtion, your problem could be anything from an error in your cascoding to a bad soldering. I don't think you have oscillation, because that tend to show up as burned components, but you never know...

[tpark]

Thanks for the reply. I do not have a software to draw schematics however mine is basically stock except for the cascode. This is basically another SK147 or SJ72 with the source connected to the drain of the main input fet, the drain is connected to the resisters as normal and the gate has the LM329DZ from ground and loaded with a 17K resistor to the regulated +. I have not added the current limiters yet and haven't hard-wired the circuit to the PS using 20 inch hookup wires with alligator clips, the output mosfets are mounted on the heat sinks. The circuit for the regulator is exactly as in Zen variations #3 and I am using 330uF caps on both the zener chain ( a 15 volt and a 13 volt in series )and the output. I have tried increasing the current through the zeners by dropping the resister from 1500 to 750 with no change. If I disconnect the .47 source resisters from both sides, the input fets bias perfectly at around 50mv and the regulators are both around 24 volts. I think I will try the other version where you use a 47.5K resistor from the PS and omit the zener chain and see how that works. I merely wanted to get rid of the ac spikes and drop some volts.

[jacco vermeulen]

Originally Posted by tpark

I have quite a few old SK147 and J72 Toshiba fets. They are limited to 15volts

Maybe I read it other than you intended, but the K147 handles 40V and the J72 does 25V, just as the 400mW and 200mW afterbirths.

[tpark]

I was looking at datasheets again and you are quite correct. I remember Erno Borbely years ago when he published his pre-amp in Audio Amateur stating not to use them higher than a drain of 15 volts and this stuck with me. On re-reading that old article I noticed that the recommendation was for low noise in his phono stage and he cascoded also to reduce the effect of the rather large input capacitance as well. I'll remove the cascodes and see if it makes a difference. I doubt noise would be an issue in an amplifier of this type. I am pretty sure my heat sinks can stand a 30volt rail.

[triode_al]

The SJ74 can handle operationally, i.e. in RMS much higher than 25V. Note that when the swing is 20V (peak) up, the gates are high too (10/60 * 25 = 4V, but that does not harm the jfet at all in 90% of F5 implementations that have a rail of 22V.

But the cascode has a different effect: reduces the gate current, gives a second harmonic structure to the input.

Tip: I would rather put the tail on the source of the jfets, or have it float in a similar way. Why I would place the tail of the cascode bias different than to earth: the 8V peak-peak is still seen by the drains. A fixed voltage cascode in the manner of the F5-C gives the jfets a very hard time!

I used a 20K/5k resistor pair for the bias. I have that in my Zen I/V implementation (to a good effect). A transistor might sound better than a cascode of two jfets. (that is also too expensive imho)
albert

[tpark]

[QUOTE=triode_al;2885149]

But the cascode has a different effect: reduces the gate current, gives a second harmonic structure to the input./QUOTE]
I don't understand what you mean by reducing the gate current. Dr. Hawkesford wrote some interesting papers about using cascodes both for the input side and also the output side and showed significant reductions in distortion. Again that was with transistors and may not apply to fets.
I agree that setting the reference gate voltage for the cascode at the source is a more desirable way of doing this and also that the pass device is probably less important and could be a transistor, tube, etc but this gets away from the Fet concept. If you use a resistor divider do you not get a modulation of the signal at the gate of the cascode device whereas a device such as a LM329DZ or Lm336-5.0 volt would keep the source to drain voltage differential more constant which would benefit the gain device and I would guess keep it more linear. I am not an engineer so probably way over my head here.
I tried removing the cascodes and with the higher drain to gate voltages the current increased however the amp worked fine. Just had to drop the RL via the trimpots. When I tried to use the regulators there was no change with the + regulator fixed at 4 volts and the - at 24 volts. The plus side shut off and the negative side conducted a small amount upseting the feedback bias. My next step is to try the capacitance multiplier approach.
I must admit I find it fascinating how Mr Pass has allowed us to use devices with a very limited class A voltage range ( eg +/- 50mv for these types ) to deal with input signals of the order of several volts all in Class A. The implications for a line amp seem clear to me.

[triode_al]

Originally Posted by tpark

I don't understand what you mean by reducing the gate current.

Hi, this is the background of what I mean:
see Le Monstre by Hiraga, 1983, in part 1. The schema shows that the configuration is similar to the F5, except the output pairs (Darlingnot) are now FETS in the F5 of a type not yet commercialized at that time imho.

Cascode input current reduction
monster27fig7 - reducing input current.gif
Getting a lower gate current improves the transparency. The gate current of a single 2SK170 will change a lot when the Vds changes - and in the F5 with the low feedback, Vds does change a lot.

cascode jfet/transistor distortion
monster27fig4a-distortion of cascode.gif
This is highly audiophile quality, good to have when applying feedback.

cascode jfet/jfet distortion
monster27fig4b-distortion of fet-fet cascode.gif
You see this is not a really 'audiophile' input combo. On feedback, higher order distortion starts to increase.

Of course the Miller effect vanishes too.
Hiraga also says that the combo of a biasing of the cascode with a zener sounds worse. (Now your "super"zener IC might sound better than a stock zener, but I just want to mention it anyway.)

Just my two pennies.
albert

[AndrewT]

Triode,
you show diagrams titled jfet/transistor and jfet/jfet.
What is the difference between these?

[triode_al]

These are the two types.

monster c ascopde types.gif
Andrew,
The jfet+cascode on the top is the winner. This nice harmonics goes for the earthed config.
So it works like this in the Zen I/V, but for the F5 the bottom of the bias circuit has to be connected to the source, sort of 'floating', as far as Hiraga's recommendations go.
albert