F5 power amplifier

> You must have what we call "adequate heat sinking" to enjoy that drift figure.

The case of the MOSFET was at 55 degC, so it was still a 30 degC rise, and the junction should be at about 80 degC, but the fact that I am biasing at 2A per FET means that it is also closer to zero tempco than the single ended version.

The heatsink itself is 0.2 K/W, and I have been reasonably successful in my thermal management and predictions.

:)

But maybe I'll get more drift with Fairchild devices. Still waiting delivery from Digikey.
We shall see.


Patrick
 
My P-to-P wired F5 is now complete!

I did not use Toshiba jfets, I used some commonly available jfets from Digikey. For the 3w resistors, I actually used 10W WW cheap sandcast resistors, the kind Radio Shack sells.

Perhaps this is the reason I see no peak in the high frequency response. 20K square waves look quite square.

So, How does it sound?

I spent some time listening to it last night, and again tonight.

It sounds amazing! It does not sound bright to me, but all the highs are very detailed......

JJ
 
The issue with the K170 is pretty high input capacitance.

This version, without any tweaking, came in with THD of a hair over 0.003% at 1W.

200kHz sq wave with no input compensation -- not quite as pretty as NP's in the May08 AX article -- the rise time for my signal generator isn't particularly fast (darn, I wish I hadn't sold the PG508).

An externally hosted image should be here but it was not working when we last tested it.
 
> If anyone wanted to order jfets from Digikey, I would go with J176's or J175's,

I would not.

J175/6 is very noisy compared to the 2SJ74s. Also it has a much larger Idss, and the Idss spread within one type is very large. You end up buying a lot to get a few. Believe me, I know. (I use them for a different application.)

I would stick to Toshiba, not only because of noise, but also because of gain (Yfs) -- compared with the 2N types. There isn't already much gain, something like 36dB NFB.

> Does your balanced F5 have the noise reduction benefits gained by using a complementary pair?

Yes, or better (because of the "X").

If you ground the intersection of the "X", you end up having 2 F5's driven as bridged amps. In each half, you have some harmonic cancellation between top & bottom half, so that is why you see almost no forth & sixth harmonics, but there is some second left, as the transistors are not truely complementary. In my case, the MOSFETs are, but the JFETs are still not. That was why I used 2SK170/2SJ74, instead of 2SK389/2SJ109, because the former are more complementary than the latter. You can try to get better cancellation for second by modifying the values of the source resistors at either the JFET (where then the Rfb also needs to be changed accordingly), or as Nelson did, at the MOSFET source, or both (as I did, if you were paying attention to my schematics). But the balanced (bridged) version will cancel the remaining even harmonics for you, if the two halves are identical (i.e. very well matched) and driven by a perfectly balanced signal with zero common mode.

The unconnected, ungrounded "X" should work like Super Symmetry. But the distortion is already lower than I can measure. Klaus did not seem to find any significant improvements in his simulation, but I do believe you do in real life with real devices, where they are never 100% matched.

Only my personal opinion / choices.
You should really ask the "One and Only". Or you build and listen and decide for yourself.

:cool:


Patrick

PS I forgot to ask you if you can post a picture of your square wave (20kHz or better still 200kHz) under reactive load (when connected to a loudspeaker) at about 1W output.
 
2A for F5 ?

Could I increase bias from 1.3A to 2A for F5 ?

Because now I had compared with my Aleph J cloned, I loved Aleph J more than F5 because its close to my 2A3 but in my remember I had adjusted bias of F5 from 1.3A to 1.75A ! I loved it !

What's problem if I change bias of F5 to 2A ? (don't tell me it's so hot !!!)

Thanks
Rubydac
 
> Could I increase bias from 1.3A to 2A for F5 ?

If you are still using +/-24V power supply, you would be running each MOSFET at 48W. Even if you have a large enough heatsink (say 0.4K/W per channel), your MOSFET junction temperature would rise to almost 100 degC. Unless of course you use water cooling to run the MOSFETs at case temperature of about 30 degC (essentially room temperature in the summer in Hong Kong).

You may ask what's wrong with 100 degC junction temperature.
Nothing, except the realibility of the MOSFETs goes down significantly. So if you are prepared to accept that the MOSFETs wil fail now and then, and you have speaker protection installed, by all means.

Of course you are also changing the "sweet spot" at which the original circuit is set by going to 2A, as the drain resistors (trimpot) of the JFets would have to be increased to increase the bias of the MOSFETs. You may of course also achieve the same by increasing JFet bias instead, by using devices with an Idss of say 10mA and keeping the drain resistors as they are (at around 700 ohm).

Or you may wish to consider using the balanced circuit I published, which runs at 16V 2A per MOSFETs, thus still keeping dissipation per MOSFET at about 32W.

I have only built and listened to my all-Toshiba balanced version, so I cannot comment on how different it may sound to single ended Fairchild. But I believe Nelson has both single ended and balanced versions, so perhaps he might care to say a few words on that.


Patrick
 
The one and only
Joined 2001
Paid Member
Re: Balanced version schematic?

Rob Alexander said:
I'm not sure if this has been already answered but I've seen a few references to a balanced version of the F5 but don't recall a schematic from Nelson, what the gain would be compared to the srandard version, whether he built one and if so how it sounds in comparison. Just curious.

Ordinarily the gain would be the same.

Balanced versions would come in various flavors.

The primary advantage of balanced output operation is more power,
since the push-pull nature of the single-ended version already has
a third harmonic character at higher power levels. Assuming that
you want to maximize the power, the first thing you are likely to do
is to double the number of output devices (2 in parallel), although
this is not essential.

Then you consider whether you simply want to "bridge" the
amplifier by taking to F5 channels and operating them across the
load out-of-phase.

Lastly, assuming that you want an input stage which integrates
what would otherwise be two separate independent input stages,
you have a couple of obvious alternatives.

The most obvious is to simply take the 10 ohm resistors to ground
and connect them to each other instead.

The other is to make an X circuit out of it.
 
> Then you consider whether you simply want to "bridge" the amplifier by taking to F5 channels and operating them across the load out-of-phase.

> Lastly, assuming that you want an input stage which integrates what would otherwise be two separate independent input stages, you have a couple of obvious alternatives. The most obvious is to simply take the 10 ohm resistors to ground and connect them to each other instead. The other is to make an X circuit out of it.


Referring to Klaus's post earlier regarding these 3 variants :

http://www.diyaudio.com/forums/showthread.php?postid=1538833#post1538833

" I simulated all three variants, the simple Bridge (the nodes in question grounded), Zinsula (nodes connected directly) and yours (nodes cross-connected only). But I couldn't find any significant change in distortion, both looking at the single-ended outputs (ref'd to GND) and the differential ouput, the third harmonic always dominated THD with its stable value. "

I wonder if you would care to comment on the merits of each of the three ?


Patrick
 
The one and only
Joined 2001
Paid Member
EUVL said:
" I simulated all three variants, the simple Bridge (the nodes in question grounded), Zinsula (nodes connected directly) and yours (nodes cross-connected only). But I couldn't find any significant change in distortion, both looking at the single-ended outputs (ref'd to GND) and the differential ouput, the third harmonic always dominated THD with its stable value. "

I wonder if you would care to comment on the merits of each of the three ?

Depends partly on the device models. If the N devices and
(separately) the P devices are literally identical which is likely the
case in your simulation, then the even harmonics cancel for similarly
and the odd harmonics are reduced by the proportion of feedback
which can also be made similar.