Power Amplifier AB class

[strawberry]

Amplifier board for amplifier repair and mod
Output power: 100W
THD+N: unknown
Slewrate: 18VuS
Bandwith: 15Hz..150kHz
Power supply: +/-10..50V
Amplifier topology: Voltage amplifier
Amplifier class: AB
Parts list: 2SA1943N,  2SC5200N,  TTA004B, TTC004B, KSA992,  KSC1845, BSS169, SSM3K15F, wima film capacitors,  1206 size chip resistors/ceramic capacitors, Piher trimmer, 0.22R low inductance KOA Speer resistors, 

[mariush]

Looks interesting but I wonder if it's really worth it when you can buy AB class amplifier chips like TDA7294 (link to datasheet and example circuit)

unless your circuit has very low thd, seems like 6$ per chip would be reasonable, when i can save pcb space and reduce component count.
plus i don't have to hunt potentially rare transistors or have to match them...

[strawberry]

Old amplifiers with STK modules... Power supply voltage is near 50V or above, wont leave any margine for TDA7294 LM3886.. 40V (recomended max)
Osciloscope and FFT give promising resaults up to 50kHz

[mariush]

It's +/- 40v ... split power supply.... so up to 80v ... you can easily use 48v AC transformers with center tap or two 24v secondary windings.

And you can wire two modules in bridge mode and power them basically from up to 80v ... see datasheet, page 13

[David Hess]

Looks interesting but I wonder if it's really worth it when you can buy AB class amplifier chips like TDA7294 (link to datasheet and example circuit)

unless your circuit has very low thd, seems like 6$ per chip would be reasonable, when i can save pcb space and reduce component count.

The integrated or hybrid solution has a lot of advantages including ease of use, size, and cost.  But a discrete solution can be higher performance and will be better about handing heat dissipation.

To just give one example, both of these designs use emitter degeneration to increase full power bandwidth which is common in audio power amplifiers however this raises the input noise by enough that it will be audible in the speaker.  A discrete design, although not this one, could avoid this to make a "noiseless" power amplifier.

plus i don't have to hunt potentially rare transistors or have to match them...

The transistors are not rare (even now) and do not need to be matched in most discrete designs.

[strawberry]

How to lower noise floor and get rid of RF interference?

[David Hess]

How to lower noise floor and get rid of RF interference?

The schematic shows RF suppression so if this is a problem, it may be a construction issue.

Lowering the noise requires removing R5 and R8 which has all kinds of implications including reimplementing the frequency compensation somehow.  The reason they are ultimately there is to improve the full power bandwidth (slew rate).  In practice I think a completely different circuit would be required.

[Zero999]

The idea of a hobby project with an IC doesn't seem like as much fun. An op-amp followed by a discrete voltage and current gain stage is often the preferred way to do this.

[strawberry]

Do amplifier can be designed without over current protection?
turns out that most of the noise comes from PC sound card headphone amplifier.

[dzseki]

To just give one example, both of these designs use emitter degeneration to increase full power bandwidth which is common in audio power amplifiers however this raises the input noise by enough that it will be audible in the speaker.  A discrete design, although not this one, could avoid this to make a "noiseless" power amplifier.

Without emitter degeneration the open loop amplification of the differential stage will be much higher, which may lead to Transient intermodulation (TIM) problems, in other words emitter degenerated differential pairs are more linear in behaviour when examined alone less demanding for the global feedback loop to get the things "right"

[dzseki]

Do amplifier can be designed without over current protection?

Of course, for example the schematic you posted is one such design.

[David Hess]

To just give one example, both of these designs use emitter degeneration to increase full power bandwidth which is common in audio power amplifiers however this raises the input noise by enough that it will be audible in the speaker.  A discrete design, although not this one, could avoid this to make a "noiseless" power amplifier.

Without emitter degeneration the open loop amplification of the differential stage will be much higher, which may lead to Transient intermodulation (TIM) problems, in other words emitter degenerated differential pairs are more linear in behaviour when examined alone less demanding for the global feedback loop to get the things "right"

Lack of linearity of the input stage does not cause TIM; lack of full power bandwidth (slew rate) does.

Adding Gm reduction of any type allows for a higher tail current and a smaller Miller capacitance (C5) without closed loop instability resulting in a higher slew rate so recovery is faster.  The easiest and most straightforward way to add Gm reduction is emitter degeneration and the added input noise is seldom a disadvantage in a power amplifier.  It is likely that without this, the full power bandwidth would not be sufficient anyway so something has to be done and it might as well be this.

[dzseki]

Why emitter degeneration affects noise so severely? I mean the typical value of emitter degeneration is a few 10 Ohms, the differential pair is usualy biased at a few milliamps. Where does the noise comes then? The induced current and voltage noise of the resistor is negligable here as I see, also without emitter degeneration the stage would have higher gain, therefore it would amplify its own noise more too.

[David Hess]

Why emitter degeneration affects noise so severely? I mean the typical value of emitter degeneration is a few 10 Ohms, the differential pair is usualy biased at a few milliamps. Where does the noise comes then? The induced current and voltage noise of the resistor is negligable here as I see, also without emitter degeneration the stage would have higher gain, therefore it would amplify its own noise more too.

You just gave the exact reason.  In the example above, the tail current is 1.8 milliamps yielding an emitter resistance of 28 ohms making the 47 ohm resistor the largest noise source.

This particular design is one of the quieter ones.  Most are much worse with the emitter degeneration being 10 times larger than the emitter resistance.

[strawberry]

Differential stage can go into weird clipping

[dzseki]

You just gave the exact reason.  In the example above, the tail current is 1.8 milliamps yielding an emitter resistance of 28 ohms making the 47 ohm resistor the largest noise source.

This particular design is one of the quieter ones.  Most are much worse with the emitter degeneration being 10 times larger than the emitter resistance.

I am still not convinced. I did a simulation on a circuit excerpt of an amplifier I am currently restoring. In this circuit the emitter degeneration is 220 Ohm and the bias is at 2.7mA (with proper current source) -see pictures.
If there is no degeneration on the emitter the simulation shows 62nV/sqrt(Hz) output noise, with 220 Ohm emitter degeneration this is only about 21nV/sqrt(Hz). This is of course because the gain of the stage is higher with no degeneration. Nevertheless this noise will propagate through the VAS regardless of the loop gain, leaving the non degenerated case noisier when everything else is the same.

[strawberry]

 when poking shcassis with screwdriver amplifier make crackling noise, zobel network is stopping from lachup oscilation but is not eliminating problem

[David Hess]

If there is no degeneration on the emitter the simulation shows 62nV/sqrt(Hz) output noise, with 220 Ohm emitter degeneration this is only about 21nV/sqrt(Hz). This is of course because the gain of the stage is higher with no degeneration. Nevertheless this noise will propagate through the VAS regardless of the loop gain, leaving the non degenerated case noisier when everything else is the same.

The gain makes all of the difference when measuring output noise.  The output noise must be divided by the gain to find the input noise.  Where the gain matters is dividing the noise contribution of the following stage which is why the gain of the differential input stage is typically low and most of the gain is provided by a following stage.

The noise from the emitter resistance is in series with the base-emitter junction voltage so it shows up as the input noise voltage directly, with the added complication that with a differential stage, the noise is actually sqrt(2) greater which is one of the disadvantages of a differential amplifier.  On the other hand, the differential configuration has common mode rejection which attenuates the noise from the tail current.

Just doing a back of the notepad calculation, I get 2.8nV/SqrtHz of input noise and a gain of 6.9 for your first example and 0.81nV/SqrtHz input noise and a gain of 85.5 for your second.  That turns into 19.3nV/SqrtHz and 69 nV/SqrtHz of output noise which is pretty close to your results.  I have no idea what the bulk resistance of a BC182 is but SPICE might.

Differential stage can go into weird clipping

The only time that should happen is when the output slew rate is exceeded preventing the feedback network from balancing the input which shows why full power bandwidth is important in an audio power amplifier.

Besides the mentioned benefit of allowing an increase in full power bandwidth, degeneration does increase linearity but in the examples I have studied, the resulting decrease in distortion was marginal and better accomplished in other ways.  For simple designs however, this is the way to go and the increased noise is hardly a disadvantage in a power amplifier except in special cases.

[strawberry]

How to calculate Long tailed pair and VAS stage amplification and slewrate?

[David Hess]

How to calculate Long tailed pair and VAS stage amplification and slewrate?

The easy way is simply the tail current into the Miller compensation capacitance of the VAS stage.  This points to why Gm reduction allows an increase in slew rate.  With a lower Gm, either less capacitance can be used for compensation or the tail current can be increased.

Of course if there is a fixed voltage amplification after the operational amplifier like in the examples I gave, it directly multiplies the slew rate at the cost of increasing the minimum closed loop gain.  This is essentially sticking a current feedback amplifier with a fixed gain on the output of a conventional voltage feedback operational amplifier.

For detailed information on this subject of tail current, compensation, and slew rate:

Designing Audio Power Amplifiers - Bob Cordell
Audio Power Amplifier Design Handbook - Douglas Self

and

AN-A The Monolithic Operational Amplifier - National Semiconductor

[strawberry]

some new great design files

[TurboTom]

You're sure Q16 and Q17 are the correct polarity? 

[strawberry]

best design for sure 

[Mechatrommer]

best design for sure 

100W + no heat sink = burnt

[mariush]

imho any design that's mostly through hole and then gives up and uses surface mount on the bottom gets disqualified.

In the example above, designer could have used a couple tiny circuit boards to place the surface mounted stuff om them and then use a couple headers to plug them into the main board (like SIP packages, RAM sticks ) ... or just use through hole parts ... so the whole design could have been single sided ... there's plenty of room on the board to do something like that.

Not to mention the lack of heatsink mounting options (Screw holes, silkscreen etc), and the weird placement of those 4 resistors at the top right by the holes