After we successfully finished the F5X-HA,
http://www.diyaudio.com/forums/pass-labs/271926-f5-headamp-12.html#post4789673
the discussion started as to what other balanced topology to choose as the next project.
The obvious answer is a balanced DAO, with a fully balanced HAGS and a DAO circlotron. A headphone amplifier topology with zero global negative feedback (ZGF), using a separate gain stage followed by a balanced power buffer. This is now being planned, and we shall report once we have more to show. But this design requires a lot of devices which are now obsolete. So not a project for anyone without a private collection of audio FETs.
But are there any other topology to consider ?
About two month ago, WalterW published a M2 Headamp, using the concept of the M2 power amp, i.e. a complementary JFET follower, followed by a step-up transformer, followed by a power buffer for headphone loads. This is also ZGF, though not balanced.
http://www.diyaudio.com/forums/pass-labs/166784-pictures-your-diy-pass-amplifier-73.html#post4751110
We wanted to follow Walter's idea to use a collection of Pass circuits to make up a balanced headphone amp with gain. But we also wanted to avoid the use of P-JFETs, as these are too expensive even though available.
So how about a M2X using an N-JFET circlotron follower as front end, with 5x gain using the same EDCOR PC600/15k a in the M2, and then another N-JFET power circlotron as output stage ?
The EDCOR has centre tap for both input and output, making it ideal for balanced operation. The primary impedance would now be 300 ohm per phase, and the secondary 7.5k.
The 2SK170BL, or the equivalent LSK170B, is still widely available for reasonable price. The average Idss is about 9mA, and the Yfs is about 25mS (equivalent to 40 ohm). To ensure reasonable low distortion, each JFET pair should not be driving more than 600 ohm load. That means the input stage should use 2 pairs to drive the transformer primary. For the output stage, we want it to have at least 150mA bias, and be able to drive 30 ohm loads. That means 20 pairs. For the 20 devices in parallel, they do not have to be Idss matched to 3%, but it would be nice to have them within 10% for equal current sharing. Does this not remind you of the “Beast of a Thousand FETs” ?
http://www.firstwatt.com/pdf/art_beast.pdf
If you want to be perfect, you can connect 20x LSK389B in parallel instead, and further connect them thermally to a common heat sink. This will make sure that both legs of the circlotron will track each other thermally to cancel out any DC drift.
And you will probably have a hard time to find a headphone amplifier circuit that will be of much lower noise.
🙂
Patrick
.
http://www.diyaudio.com/forums/pass-labs/271926-f5-headamp-12.html#post4789673
the discussion started as to what other balanced topology to choose as the next project.
The obvious answer is a balanced DAO, with a fully balanced HAGS and a DAO circlotron. A headphone amplifier topology with zero global negative feedback (ZGF), using a separate gain stage followed by a balanced power buffer. This is now being planned, and we shall report once we have more to show. But this design requires a lot of devices which are now obsolete. So not a project for anyone without a private collection of audio FETs.
But are there any other topology to consider ?
About two month ago, WalterW published a M2 Headamp, using the concept of the M2 power amp, i.e. a complementary JFET follower, followed by a step-up transformer, followed by a power buffer for headphone loads. This is also ZGF, though not balanced.
http://www.diyaudio.com/forums/pass-labs/166784-pictures-your-diy-pass-amplifier-73.html#post4751110
We wanted to follow Walter's idea to use a collection of Pass circuits to make up a balanced headphone amp with gain. But we also wanted to avoid the use of P-JFETs, as these are too expensive even though available.
So how about a M2X using an N-JFET circlotron follower as front end, with 5x gain using the same EDCOR PC600/15k a in the M2, and then another N-JFET power circlotron as output stage ?
The EDCOR has centre tap for both input and output, making it ideal for balanced operation. The primary impedance would now be 300 ohm per phase, and the secondary 7.5k.
The 2SK170BL, or the equivalent LSK170B, is still widely available for reasonable price. The average Idss is about 9mA, and the Yfs is about 25mS (equivalent to 40 ohm). To ensure reasonable low distortion, each JFET pair should not be driving more than 600 ohm load. That means the input stage should use 2 pairs to drive the transformer primary. For the output stage, we want it to have at least 150mA bias, and be able to drive 30 ohm loads. That means 20 pairs. For the 20 devices in parallel, they do not have to be Idss matched to 3%, but it would be nice to have them within 10% for equal current sharing. Does this not remind you of the “Beast of a Thousand FETs” ?
http://www.firstwatt.com/pdf/art_beast.pdf
If you want to be perfect, you can connect 20x LSK389B in parallel instead, and further connect them thermally to a common heat sink. This will make sure that both legs of the circlotron will track each other thermally to cancel out any DC drift.
And you will probably have a hard time to find a headphone amplifier circuit that will be of much lower noise.
🙂
Patrick
.
Attachments
Last edited:
You have all the bit to build one, I am sure. 😉
Epecially if you have a high impedance phone, the number of FET can be reduced.
Patrick
Epecially if you have a high impedance phone, the number of FET can be reduced.
Patrick
The circuit in post #1 requires 4x floating PSU per channel. That can be somewhat prohibitive.
Luckily, Mr. Broskie comes to the rescue.
By burning more power and adding 2x resistors and 2x current sources per circlotron, one can reduce each channel to dual rail fixed supplies.
Since current draw is constant, one may even use one supply for both channels.
Circlotron Circuits
Here I assume the front end bias to be 18mA, and the power stage 180mA.
If different the values of the added resistors and CCS's need to be adjusted to suit.
Patrick
.
Luckily, Mr. Broskie comes to the rescue.
By burning more power and adding 2x resistors and 2x current sources per circlotron, one can reduce each channel to dual rail fixed supplies.
Since current draw is constant, one may even use one supply for both channels.
Circlotron Circuits
Here I assume the front end bias to be 18mA, and the power stage 180mA.
If different the values of the added resistors and CCS's need to be adjusted to suit.
Patrick
.
Attachments
Last edited:
Interesting proposition!! The headamp with 98 Jfets!!
Do we need to have all fets per channel within 10% Or could we get away with matched total current draw? This way I could add the Idss values of 20 jfets to a close total value. All within about 15% matching between them.
Do we need to have all fets per channel within 10% Or could we get away with matched total current draw? This way I could add the Idss values of 20 jfets to a close total value. All within about 15% matching between them.
I counted 88 FETs ?
The front end quad wants to be well matched, more like 3%, and Idss sum <1%.
For the 40x per channel at the output, 15% would be fine, but you do want the um to be <1% for minimum DC.
And if you want lowest distortion, I would consider a better transformer, such as Sowter 9062.
I am not expert with Tx, so you should do some more research.
Patrick
The front end quad wants to be well matched, more like 3%, and Idss sum <1%.
For the 40x per channel at the output, 15% would be fine, but you do want the um to be <1% for minimum DC.
And if you want lowest distortion, I would consider a better transformer, such as Sowter 9062.
I am not expert with Tx, so you should do some more research.
Patrick
Last edited:
BTW one can replace 2x 2SK170BL with one 2SK369V, though the latter is as difficult to get.
But 88 is a number that brings luck, and certainly not 44.
🙂
Patrick
But 88 is a number that brings luck, and certainly not 44.
🙂
Patrick
That's what I get for math late at light after a bottle of wine! 88 it is indeed!
I have some well matched quads (Idss at room temperature) for the input stage and then I have about 100 BL grade ones to try to match for the output!
Jfet trading by Idss anyone? 😀
Would the BF862 be a good candidate for an smd version? We could use top and bottom layers to get very small circuit paths and also not run out of old jfets!
You need to cascode the BF862 to Vds 5V or so.
See datasheet Fig. 9.
And then you need to add 5V to each of the power supplies.
BTW if you have 50 ohm phones, you can reduce the number of FETs at the output stage to say 12x parallel.
Patrick
See datasheet Fig. 9.
And then you need to add 5V to each of the power supplies.
BTW if you have 50 ohm phones, you can reduce the number of FETs at the output stage to say 12x parallel.
Patrick
Last edited:
- Status
- Not open for further replies.