My first amplifier design

[jaycee]

Well, here it is:

http://www.darkmatter.myby.co.uk/cascodevas.png

(I didn't make it an image as it's huge - LTSpice looks nasty at smaller resolutions)

I've never really designed an amplifier before, and haven't had a lot to go on, just some articles on the internet, looking at lots of designs, and tinkering in SPICE, so I'm looking for some expert opinions here.

A few points:

* 2SC2240/2SA970 will most likely be replaced with 2SC2547E/2SA1085E instead as they are lower noise. 2N5401/2N5551 will probably be used for early testing as they are cheap!

* BD139/140 are on the edge of their ratings, I will hopefully use 2SC2911/2SA1209 instead.

* MJL21193/MJL21194 will probably be changed for MJL4281/MJL4302 - I didn't use the SPICE models for those as they are apparently wrong.

One question I do have, there is a large turn-on thump due to the current sources in the input LTP's... is there any way I can calm this down as it takes about 2 seconds to settle according to the simulator.

It's designed for 100W at 8 ohms with a 1V RMS input (gain 30.46dB)

[sam9]

The best thing to do is build it and see what happens.

The output section looks solid -- EF topology has is pretty trouble free once you get the bias setting stable. However, I question whether you need a tripple unless you plan to parrallel more output devices in a subsequent version.

I notice the base resistors on the differential pairs. I'm not sure if I see this often. What is the intent/purpose? Regarding the 970/2240s in the VAS cascode -- I've had a situation in the past where high hfe transistors in those positions contributed to instability in the VAS. Substituting a considerably lower hfe transistor improved the situation with no increase in measured distortion (ltspice predicted otherwise).

In any case build it. Unless you are very, very experienced the odds of getting it right the first time are slim no matter how much simulating you do. Just get it stable and functional at first even if you have to accept higher THD than you want. It's really hard to opyimize an amplifier that keeps trying to commit suicide.

[jaycee]

Originally posted by sam9
The best thing to do is build it and see what happens.

Hopefully I shall be able to do this soon...

Originally posted by sam9
The output section looks solid -- EF topology has is pretty trouble free once you get the bias setting stable. However, I question whether you need a tripple unless you plan to parrallel more output devices in a subsequent version.

Basically I'm being influenced by the Leach amplifier... the Triple Darlington seems to have a lot of advantages, and it had lower crossover distortion when I tested it in LTSpice. I stayed away from CFB after doing a smaller amplifier and finding that the CFB is murder to keep stable.

Originally posted by sam9
I notice the base resistors on the differential pairs. I'm not sure if I see this often. What is the intent/purpose?

Again see Leach amplifier. My thought on this was that it provides some base stopper isolation between the two halves of the circuit.

Originally posted by sam9
Regarding the 970/2240s in the VAS cascode -- I've had a situation in the past where high hfe transistors in those positions contributed to instability in the VAS. Substituting a considerably lower hfe transistor improved the situation with no increase in measured distortion (ltspice predicted otherwise).

Noted. I have seen a mix of things done with cascodes... some use two small signal devices, some a small signal and medium power device, and some use dual medium power devices.

Originally posted by sam9
In any case build it. Unless you are very, very experienced the odds of getting it right the first time are slim no matter how much simulating you do. Just get it stable and functional at first even if you have to accept higher THD than you want. It's really hard to opyimize an amplifier that keeps trying to commit suicide.

The big problem is that as of yet, I don't have a 'scope, so if it starts oscillating or doing other wierd things I have no way to track it down. This is why I'm keeping my eye on eBay for a good 50MHz dual channel 'scope

[AndrewT]

Hi Jaycee,
a twin beam scope would be even better. No chopping between channels because it has two completely separate channels all the way from input to the spot on the CRT. It might then have twin channel on each beam.

[djk]

I have found that the Leach sounds better with resistors for the current sources for the diff pairs rather than an active current source.

It also sounds better with less gain in the diff pairs as in the V4.5

It also sounds better with the nested FB loop to the pre-drivers

The main reason the Super Leach has a cascoded VAS is for power handling, it is not needed for smaller versions.

"I have been asked why I did not use cascode stages (common emitter followed by common base) for the second stage. The cascode stage minimizes the Miller effect and exhibits a lower input capacitance than the common emitter stage. It is used in applications where bandwidth is to be optimized. A feedback amplifier, however, must have a dominant pole to set its gain bandwidth product. It can be shown that the optimum method to set this pole is to place a capacitor from output to input of the highest gain inverting stage. If I had used cascode amplifiers in the second stage, a capacitor would have to be placed from the output to the input of each to set the dominant pole. Then the cascode stage would have the same bandwidth as the common emitter stage. A cascode stage cannot be used to improve the bandwidth of a feedback amplifier if the gain bandwidth product is to be held constant. If the gain bandwidth product is not intentionally limited to some reasonable value, the amplifier would not be stable. That is, it would oscillate. "

[sam9]

I'm keeping my eye on eBay for a good 50MHz dual channel 'scope

I don't know your budget but consider this
http://www.elexp.com/tst_s500.htm
It's not the equivalent of an HP ot Techtronics but it is a lot of capability for the price.

quote:
Originally posted by sam9
The output section looks solid -- EF topology has is pretty trouble free once you get the bias setting stable. However, I question whether you need a tripple unless you plan to parrallel more output devices in a subsequent version.

Basically I'm being influenced by the Leach amplifier... the Triple Darlington seems to have a lot of advantages, and it had lower crossover distortion when I tested it in LTSpice. I stayed away from CFB after doing a smaller amplifier and finding that the CFB is murder to keep stable.

I wan't thinkinf Tripple-EF vs CFB. I agree CFB can be a pain - similar experience to you. I was think of a simpler EF just drivers and output devices.

[jaycee]

Originally posted by AndrewT
Hi Jaycee,
a twin beam scope would be even better. No chopping between channels because it has two completely separate channels all the way from input to the spot on the CRT. It might then have twin channel on each beam.

This is what I meant, I think... "dual trace" is what I meant, but yeah, one capable of displaying 2 traces at once.

Originally posted by sam9
I don't know your budget but consider this
http://www.elexp.com/tst_s500.htm
It's not the equivalent of an HP ot Techtronics but it is a lot of capability for the price.

I've looked at PC based scopes before, the Velleman one isn't too bad from what I've heard, and BitScope is interesting... but they're still out of my price range compared to an analogue 'scope from eBay which can go for around the £100-130 mark.

[darkfenriz]

Hi jaycee
It looks pretty standard and surely will work nice.
One thing I've noticed is you have low feedback factor, so you will have to run minimum 50mA output stage bias to keep Xover distortion low.
Another thing is why you are using miller compensation combined with cascode. It deosn't make sense to me, as cascode is made to (besides of dissipation issues) eliminating miller effect. Why don't you use a complementary Vbe multiplier with midpoint to the ground thru 23.5k R instead of these two 47k. You can also put a cap parallel to this 23.5k R to place a dominant pole instead of miller. hm...?
regards

[jaycee]

Originally posted by darkfenriz
One thing I've noticed is you have low feedback factor, so you will have to run minimum 50mA output stage bias to keep Xover distortion low.

Are you referring to the global feedback of 30.46dB or the fact that there is no local feedback between the drivers/predrivers and the outputs?

Originally posted by darkfenriz
Another thing is why you are using miller compensation combined with cascode. It deosn't make sense to me, as cascode is made to (besides of dissipation issues) eliminating miller effect.

I was getting ringing on the edge of a 1KHz square wave without compensation. I am wondering whether using RC filters would be better as I have seen in a number of designs

Originally posted by darkfenriz
Why don't you use a complementary Vbe multiplier with midpoint to the ground thru 23.5k R instead of these two 47k. You can also put a cap parallel to this 23.5k R to place a dominant pole instead of miller. hm...?

I'm no big expert. So i'm not sure what you mean here.

[darkfenriz]

30dB of feedback factor??
Maybe I looked too shallow
I mean this: