diyAudio

diyAudio (http://www.diyaudio.com/forums/)
-   Headphone Systems (http://www.diyaudio.com/forums/headphone-systems/)
-   -   my take on the szekeres (http://www.diyaudio.com/forums/headphone-systems/161306-my-take-szekeres.html)

geekysuavo 14th February 2010 09:03 PM

my take on the szekeres
 
4 Attachment(s)
i recently built an actively-loaded (CCS) szekeres headphone amp to drive my hd595 50-ohm headphones, to much success, documented here.

so, since i liked the sound of the ac-coupled szekeres, i decided i'd take the design a bit further:
- replace the lm317 ccs with an irf610, buffered with an npn bjt.
- increase the quiescent current draw (again) to 550mA.
- add a low-pass filter to the input to smoothly roll off HF.
- increase output capacitor (again) from 4700uF to 8000uF.

while none of these modifications are completely original, i think they're all worth adding. i've attached the simulation results from spice.

the irf610 dissipates around 6W and the irf640 around 2W, if my calculations are correct, and spice predicts 0.018% THD for a 1Vpp, 1kHz input. (lower than that simulated for resistor-loaded and lm317-loaded szekeres) frequency response seems adequate.

of course, i'm looking for feedback from anyone who has comments or corrections to the design. your insights are welcome, as i'll be the first to posit that i'm not the most experienced circuit designer. so, what do you think? :)

~ brad.

miket6000 17th February 2010 08:19 AM

It looks good to me, the only thing I question is why the output cap is so large (8000uF driving a 50R load gives a cut off frequency of about 0.4Hz). The extra inductance which comes with such a large cap, while small, does have a detrimental effect. I do note that you've bypassed the cap but this is a bandaid for a problem which can be mitigated by using a smaller cap, say 1000uF and still have an excellent low frequency cutoff of around 3Hz. Worth noting also is that the input cap and the bias resistors create a lowpass filter of around 1.4Hz anyway.

A possible suggestion also, while I've not done the math to figure out where your DC bias point is myself, from your description of the power dissipation in the two FETs I'd expect your DC bias voltage to be somewhere around 4V. This will allow an asymmetric maximum voltage swing. While the 3V or so of negative swing is plenty for most headphones, why not move your output bias up to ~7V to give yourself more headroom? An additional benefit of this is the more balanced power dissipation which should make heat sinking easier.

I hope this is of some use to you.


Mike.

geekysuavo 23rd February 2010 10:41 PM

Quote:

Originally Posted by miket6000 (Post 2088346)
... the only thing I question is why the output cap is so large ... The extra inductance which comes with such a large cap, while small, does have a detrimental effect.

hmm, an interesting observation on the cap inductance at that size... never considered that. i suppose that inductance due to parallelling multiple (5-8x1000uF, or 2-3x3300uF)elna silmic ii's together would only aggravate the problem, right?

Quote:

Originally Posted by miket6000 (Post 2088346)
... this is a bandaid for a problem which can be mitigated by using a smaller cap, say 1000uF and still have an excellent low frequency cutoff of around 3Hz.

i used a 4700uF panasonic FC in my first szekeres, and it seems to be ok. what would parasitic inductance 'sound' like if it did exist?

Quote:

Originally Posted by miket6000 (Post 2088346)
Worth noting also is that the input cap and the bias resistors create a lowpass filter of around 1.4Hz anyway.

but a 4.7uF input cap and 100k bias shunt resistor yield a -3dB corner of 0.4Hz.

Quote:

Originally Posted by miket6000 (Post 2088346)
... I'd expect your DC bias voltage to be somewhere around 4V. ...

makes sense. ok, so i changed the bias series resistor to 25k, the ccs collector resistor to 1.5k, and the rail voltage to 18V and the simulator gives me a gate bias of 14.4V and a source bias of 7.2V. thanks!

thanks for the feedback! :)
~ brad.

miket6000 23rd February 2010 11:24 PM

Using multiple output caps in parallel has the effect of reducing the ESL which actually helps you. The effect of parasitic inductance can be two fold, the most obvious is that it will create a lowpass filter attenuating some of your higher frequencies. Usually this occures well above the audiable range, though that never stopped audiophiles from worrying about things. The second and possibly more damaging possibility is oscillation in extreme cases. This is expecially a problem when you parallel caps. A parallel LC circuit has a minimum impedance at its resonant frequency and if you have any noise close to this frequency you can end up with oscillation. Most of the time you can ignore this and you'll never have a problem, but it's something to keep in mind, especially if you start having strrange problems.

For your input cap, you need to include the effect of R4 which, for AC signals, is effectively in parallel with R3. This is how I came to the 1.4Hz figure.

Always happy to help.


Mike.

geekysuavo 23rd February 2010 11:41 PM

ah, ok the 1.4Hz makes sense now too. my plan is to power this using a salas shunt regulator, so noise shouldn't be a problem. depending on how the rest of my simulations go, i'll be parallelling 5-8 1000uF elna silmic ii 'lytics for each channel, so i'll check out the specs on those caps.

thanks again for all the input!

~ brad.

geekysuavo 26th February 2010 02:28 PM

perhaps an even better idea than simply springing for higher quality components and better bias points: has anyone tried this circuit with a dc servo? if so, what difference in sound did you observe?

i want to keep the amp musical and not venture into the analytical...

thanks,
~ brad.

ashok 26th February 2010 02:51 PM

A lot of amps use DC servos. Some people on this forum despise servos! So it's application is very critical and if not done right it degrades the sound.
I haven't seen what needs to be done as points 1,2,3 etc. to make sure it is inaudable. It must be to do with the right time constants of the dc filter and the loading effects of the filter circuit.

Search for 'dc servo' on this forum and you might come across several experiments and opinions. Remember that the FINAL opinion is always yours and it CAN be different from what others think. So DO experiment ! AND please report your view point even if it differs from 99% of the others !

geekysuavo 26th February 2010 10:51 PM

2 Attachment(s)
as a sanity check, i threw together a basic breadboarded circuit. one irf640 with 10 ohm source resistor to ground. B+ was 13.8V. i used an ad823 with integrating R=470k and C=0.47uF (tau=0.22s) to set the gate bias through a 100k resistor. similar to the attached image found on the szekeres amp addendum page at headwize, but without the 2.2k resistors and extra rc filter before the integrator.

works as it should. the scope shows that the source voltage tracks the voltage set using the resistive divider and the fet is biased well into class a.

using a 470uF (small, yes) lytic on the output, a listening test confirmed the amp works in such a configuration. (yeah, kind of defeats the point of the servo, but i don't have a split psu.) it sounded about as good as a resistor-loaded, overbiased, high input capacitance szekeres can. ;)

~ brad.

geekysuavo 26th February 2010 10:52 PM

1 Attachment(s)
forgot to post the jorgenson servoed szekeres circuit, which i seem to recall reading somewhere that it has only been simulated, never tested. (until today, sort of.)

~ brad.

geekysuavo 26th February 2010 11:29 PM

wait a second...
 
Quote:

Originally Posted by geekysuavo (Post 2100429)
works as it should.

actually, now that i (finally) got around to measuring, there's almost a 1V difference (source 8.93V, resistor divider 9.73V) between the voltage divider and the source, and i'm not sure why... any suggestions from more experienced folk?

[edit: could it be the fact that i'm running both the amp and the servo off the same rail voltages? though the ad823 is a rail-to-rail opamp...]

~ brad.


All times are GMT. The time now is 10:28 AM.


vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2014 DragonByte Technologies Ltd.
Copyright 1999-2014 diyAudio


Content Relevant URLs by vBSEO 3.3.2