Tír Na HiFi

ill be bringing a class a mosfet prototype version for comparison that i think sounds sweet too
should be fun

I'll pop over for a listen on Sat evening, esp to hear the Class A prototype!!
KM

great, thanks ken.

t

Sound OK for me, will try to come then. :) Thanks.

A didn't listen it much in Gerry's place, coz the Quad's didn't suit the amp, but that's my ears, so don't get me serious. :)

You can drop me a PM with directions. Cheers.

nige2000 wrote: ↑Thu Jun 22, 2017 9:27 pm ill be bringing a class a mosfet prototype version for comparison that i think sounds sweet too
should be fun

Is it your own design or based on something of Nelson Pass? I'm curious as I've been playing with a classA SE headphone amp of late with a Toshiba MOSFET in it which on paper has excellent credentials. Sounds marvelous too....

Yes it's basically a pass f5 with jfet signal buffer in front
I did get some Toshiba fets to try in final revisions
Too expensive to risk while still mucking about with design
I like it but it's massive, hot and expensive
All going well I'll go mono with it that will keep it relatively cool
I just get the feeling mono will be much better

tell me more bout the headphone amp

The headphone amp is very low power in its standard form (100mW peak) but could be extended (paralleled up) to increase the output. The interesting aspects are that its using a single high transconductance FET (not the typical Toshiba FETs designed for audio) and that its configured as a shunt, across the load. The upshot of this is - no variations on the power supply, at least to the extent that the current source load has high enough output impedance. Put another way, this amp should have very high PSRR meaning there shouldn't be need for heroics as you've done with your LM-based beauty here.

I reckon this approach could be scaled up to enough output power for a fairly efficient speaker (say 10W peak) but don't have any such speakers in my possession which would drive me to build such an amp.

I took a quick look at the F5, looks to be push-pull in the output so not quite the same beast as I'm doing (single-ended).

yea that sounds really good, my knowledge of class a and fets is very basic (although i get lucky sometimes),
is that a similar effect or byproduct of setting a really high idle bias ?

yea the f5 is push pull
i was trying to get some simplified pcbs designed with an integrated pre buffer to put into some overkill heatsinking
any thoughts on a way to improve

im sure ill try single ended yet, theres supposed to be a certain magic to it?

You're right about the bias, you do have to set it to the max output current which means the efficiency's very low, 25% at best. Hence heatsinks need to be really substantial. Best way is to get the dissipation spread out over a few paralleled devices rather than concentrate it in one or two big die b*ggers.

I see people on DIYA talking about the 'SE magic' and attributing it to 'harmonic profile' but I'm somewhat sceptical that this is the reason for why they sound good. Also I think its a waste to go SE classA but then not arrange the active device(s) as shunts so the power supply draw is invariant. Take xrk's latest headphone amp, he's put the load in parallel with the source resistance, which means non-constant supply demands. If he just inverted the output stage (swapped N-chan FET to P-chan) he could fix this up I reckon.

Looking at the classic Toshiba FETs which are broadly approved by amp gurus (2SK1530) I am wondering if better performance couldn't be got from running multiple paralleled FETs. I showed a way of doing this on xrk's headphone amp thread but he maintained it needed FET matching. It doesn't though - I have a prototype slightly higher powered (about 250mW) headphone amp with 5 sets of paralleled FETs (these are tiny, SOT-23 package) and it works a treat without FET matching because I don't parallel at DC, only via caps.

abraxalito wrote: ↑Fri Jun 23, 2017 2:37 pm it works a treat without FET matching because I don't parallel at DC, only via caps.

right thats were you lost me how do you parallel at caps?
any example fets for parallelling?