Output stage transistor type

[mcd99]

Beyond a certain value, gate stoppers alone may not stabilize a vertical MOSFET, since there are parasitic capacitances on the gate side of the gate stopper that may bypass the gate stopper in creating an oscillator topology. Damping of the gate circuit, right at the gate pin, is important.

Hi Bob,

Still toying with the idea detailed in your MOSFET Power Amplifier with Error Correction. This is the one where you use coax with a 10R base stopper and ferrite bead.

Have an amplifier stable using the same values for gate stoppers and zobels as the amplifier in the above paper. Does the capacitance of the coax have a damping effect? Or can you just drop the 10R gate stopper with ferrites straight in and remove the gate zobel?

Paul

 

[tvrgeek]

Got it stable. Changed the emitter resistor on the VAS clamp to 200 Ohms. Why it was the critical path I do not know.

Anyway, on distortion with respect to output types. I modeled variations on a theme.
As close as possible to the same IPS and VAS, only tweaking for stability:
20Khz, 1-9 harmonics, @ 50W out

All BJT, single output pair .000142% but the drivers loose control of the outputs at 5 Ohms.
Two output pairs .000335% Obvious rail sticking but drivers OK
Lateral MOSFET drivers .000423% Less over head but better clipping recovery
Lateral drivers, HEX outputs .000497% pretty good behavior.
I did not model lateral outputs or BJT drivers and MOSFET outputs.

So, I conclude the output distortion is basically not the driving factor. All can be made very low. Secondary issues like low ohms and clipping behavior are more important.

A few other observations:
Seperating the IPS and VAS current sources helped a lot at clipping. Not much under normal drive.
IPS BJT/LED seemed a tad more stable and only a tad more distortion. Choose your poison.
For this little amp, the flying catch diodes work like a charm. Soft recovery ultra fast Schottky simmed the cleanest.
Even for a little amp with big outputs, two pairs really shows it's worth under stress.
If I were to build either MOSFET version, I would go a little higher Vcc to regain the overhead.

I think I am at the point I can work on a board design. One board should support any of the iterations. Then I'll decide if I am building some based on what some boards cost. Now I need a chassis too as the one I was going to use I am going to stuff a gain-clone into. Better fit for my need. Maybe an Adcom 535 or 545 chassis.

 

[tvrgeek]

I still need to understand and add provisions for gate stoppers and/or Zobels. What I learned on the DH-120 is reality is a lot touchier than SPICE.
Anyway, some .asc files for overall criticism.

Of course, basic DC, temp sense, and on/off muting missing, provisions for DC null etc.

 

[Willafizz]

Well, good luck with the practical part. My situation may be a bit easier as I am building active speaker where the amplifier is quite close to the drivers. In other words, once characterised, the load will be a known and stable quantity. The other thing that has occurred to me is that it may be simpler to optimise an amplifier that is restricted to a narrower frequency band. At the moment, I am working on the PSU. It all takes time!

 

[ATAUDIO]

While I’m in the process of regurgitating, here is little more about mosfets. Vertical type generally have much higher Gm and thus higher Ft. This can be a problem in that the frequency operating limit must be set externally so they do not oscillate and cause RF ringing. They are designed for switching and low Rds on is the aim. Generally there are a few types of vertical power mosfets, trench, hex (plural well), or planer stripe (single well). (photo 1)

Cgs stays dominate with trench fets. Trench fets are not suitable for analog power usage because they tend to ‘hot spot’ very much like BJT’s do but for different reasons. This limit’s the reactive power they can handle. Stay away from trench fets for this purpose, they really are switching transistors. For the other types, Cds increases exponentially at Vds saturation. This means it takes more di/dt to pull the device out of saturation, ie a stronger driver stage. Hex fets and planer stripe fets are suitable for analog applications and tend to be very rugged. I have been impressed by the abuse planer stripe fets can take. However, they do show a tendency to have a greater dependence of Gm upon Vds much more than the plural well devices (photo 2). Vds breakdown determines how wide the channels are, and this affects Pd. When choosing a fet, you want to balance Vds breakdown with Gm. Generally, higher Vds max = lower Gm. Higher Gm = higher capacitances. As per the second photo in post 16, choosing N-ch and P-ch devices with similar Gm is important. This in itself limits the choices available. Sifting through the other criteria along with what is available price wise will limit even further the optimum choices.

Going a step back to the different MOSFETS technologies, we have Laterals and then Verticals. Vertical come in hex an planar stripe flavours, that seem both to be quite well usable. Trench, according to many are not.
Anybody have ever tried those http://ixapps.ixys.com/PartDetails.aspx?pid=4831&r=1 L2 devices from IXIS?

They claim specific design for Linear applications, for once. For what I understood they are an hex-like design however. They show huge (in the tens of pF range) Cgs, but, again that should be not a problem with a proper driver.
Also they seem to lack of P channel types, but, again, this seems a general problem with MOSFETS, and I have nothing against quasi-comp.

 

[Bob Cordell]

Hi Bob,

Still toying with the idea detailed in your MOSFET Power Amplifier with Error Correction. This is the one where you use coax with a 10R base stopper and ferrite bead.

Have an amplifier stable using the same values for gate stoppers and zobels as the amplifier in the above paper. Does the capacitance of the coax have a damping effect? Or can you just drop the 10R gate stopper with ferrites straight in and remove the gate zobel?

Paul

Hi Paul,

I think there were two things at work with the coax approach. The use of coax strongly suppresses the inductance of that interconnect. In my protoype, I recall that there was perhaps 2-3 inches from the drivers to the MOSFETs, so inductance would have been a concern. Secondly, the capacitance of the coax, in combination with the gate stopper, probably does result in some HF damping.

I have generally used the gate Zobel approach.

Its worth noting that the inductive nature of the output impedance of the emitter-follower driver may play a role in parasitic oscillation creation as well. The gate Zobel probably helps damp this impedance as well.

I also think it is wise to use a bit of HF R-C decoupling in the rail supply path from the output stage drains to the driver collectors. We want to suppress HF sneak paths. This can also be beneficial to the stability of BJT Triples as well.

Cheers,
Bob

 

[unclejed613]

Often wonder, whether V FETs setup to be very fast could allow an amp to be designed for a higher ULGF. This could give more feedback and reduce high frequency distortion. Would BJTs still have the edge?

If we bring error correction like Hawksford to the table are BJTs, in your opinion, still better?

So far in my limited experience, BJTs and V-FETs can sound good but L-FETs have been a little disappointing.

no, it wouldn't have much of an effect on open loop gain, but it does have an effect on stability. if you read Linear Technology's AN-47, specifically the section on amp stability, you will find that if your diff/vas is very fast, but the output stage is very slow, the stability problems increase.

that said, however, i've noticed that FET amps tend to have much larger compensation caps than BJT amps.

 

[mcd99]

Hi Paul,

I think there were two things at work with the coax approach. The use of coax strongly suppresses the inductance of that interconnect. In my protoype, I recall that there was perhaps 2-3 inches from the drivers to the MOSFETs, so inductance would have been a concern. Secondly, the capacitance of the coax, in combination with the gate stopper, probably does result in some HF damping.

I have generally used the gate Zobel approach.

Its worth noting that the inductive nature of the output impedance of the emitter-follower driver may play a role in parasitic oscillation creation as well. The gate Zobel probably helps damp this impedance as well.

I also think it is wise to use a bit of HF R-C decoupling in the rail supply path from the output stage drains to the driver collectors. We want to suppress HF sneak paths. This can also be beneficial to the stability of BJT Triples as well.

Cheers,
Bob

Hi Bob,


Thank you for the useful information. Your posts are always educational.

Is there an argument if going for a very fast output stage in adding zobels to the driver outputs?

Going to add the decoupling to the drivers as you suggest. I think there is still a burst of oscillation going on in my amp. It is very low level 30mV and high frequency. My slow scope has problems displaying it properly.

no, it wouldn't have much of an effect on open loop gain, but it does have an effect on stability. if you read Linear Technology's AN-47, specifically the section on amp stability, you will find that if your diff/vas is very fast, but the output stage is very slow, the stability problems increase.

that said, however, i've noticed that FET amps tend to have much larger compensation caps than BJT amps.

Interesting observation. May be there is a general theme of large gate stoppers which lead to a slow MOSFET output stage. Then maybe the preceding stage is slowed down as a remedy to instability.

Will have a read of the paper above.

My thinking is that if you could speed up the OPS then you could increase the frequency of dominant pole compensation for given stability. Then for a given higher frequency point say 20Khz you could end up with more loop gain. Also, thought that amplifier stability can be traded for performance. If you can reduce the phase shift from the output stage pole then you have more stability to trade.

Paul

 

[tvrgeek]

I think I have the definitive answer on transistor type.

Ready...


One you can actually buy.

OK, laterals seem to only come from Exicon and Alfet. Basically you have your choice of a single die 8A or a two die 16 amp. OK, but I can only get either from specialty suppliers in the UK, so not only are they 20 times the cost of a BJT, the shipping doubles that.

All my work for a Lateral driver, well that is limited to the 8 A big boys now. No smaller ones are in the offering I can find. Still, an ALFET may be a suitable driver for some HEXFETS. No, I am not interested in buying counterfeit NOS off e-bay.

Trench are no good due to the hot-spot issue, so that leaves us with IRF HEXFETS as the only other MOSFET answer.

 

[jacco vermeulen]

OK, laterals seem to only come from Exicon and Alfet.

Get real : MOSFETs for amplifier | Renesas Electronics America

 

[Willafizz]

Trench are no good due to the hot-spot issue, so that leaves us with IRF HEXFETS as the only other MOSFET answer.

Regarding the OPS drivers, I finally settled for a set of 2SK1530/2SJ201's. I will have a number of them which will allow me to do some matching. One thing that attracted me was the Gm spec which seems to be roughly the same for the P and N channel ones. Unless I made a mistake, it seems that the IRFs differ more strongly in this respect. I got the 2SK/2SJs from Ampslab.

I agree that the cost of shipping is a pain. Ampslab managed to get a good price but that is clearly not always the case!

 

[mcd99]

Trench are no good due to the hot-spot issue, so that leaves us with IRF HEXFETS as the only other MOSFET answer.

I'm inclined to agree with you. Especially when you consider cost and availability. That's why I've settled on IRFP240/9240 within Hawksford error correction. You can certainly get some good results with that combination.

 

[Bensen]

Have a look at irf9610/irf710. These look like a good complementairy pair for drivers.
I believe intersil specifies the Gm curve of the irf710 ont hier datasheet's.

 

[CBS240]

I think I have the definitive answer on transistor type.

Ready...


One you can actually buy.

..................

Trench are no good due to the hot-spot issue, so that leaves us with IRF HEXFETS as the only other MOSFET answer.

Fairchild makes the planer stripe type fets, quite rugged and available at reasonable prices in both polarities. There are other semi manufacturers that make stripe fets now, but Fairchild's line of stripe fets is labeled as Q-fet. FQP...... label is for TO-220, FQA...... label is for TO-3P. The data sheet will refer to their Q-fet line of mosfets. If the Fairchild datasheet does not say Q-fet then it may be a plural well device or a Trench type. Most Trench fet datasheets state that is the technology used but not all.

 

[unclejed613]

I think I have the definitive answer on transistor type.

Ready...


One you can actually buy.

OK, laterals seem to only come from Exicon and Alfet. Basically you have your choice of a single die 8A or a two die 16 amp. OK, but I can only get either from specialty suppliers in the UK, so not only are they 20 times the cost of a BJT, the shipping doubles that.

All my work for a Lateral driver, well that is limited to the 8 A big boys now. No smaller ones are in the offering I can find. Still, an ALFET may be a suitable driver for some HEXFETS. No, I am not interested in buying counterfeit NOS off e-bay.

Trench are no good due to the hot-spot issue, so that leaves us with IRF HEXFETS as the only other MOSFET answer.

many of the pro audio companies use HEXFETS in their fet amplifiers, such as Marshall's ValveState amps, etc, use HEXFETS for the amp output stage.

 

[Bob Cordell]

Hi Bob,


Thank you for the useful information. Your posts are always educational.

Is there an argument if going for a very fast output stage in adding zobels to the driver outputs?

Going to add the decoupling to the drivers as you suggest. I think there is still a burst of oscillation going on in my amp. It is very low level 30mV and high frequency. My slow scope has problems displaying it properly.

Paul

Hi Paul,

I have never tried a MOSFET driver, nor experimented with Zobels on any kind of driver, so I really don't know if there would be an advantage. I'm guessing it would be less needed than for HEXFETs in an output stage.

Cheers,
Bob

 

[Walkalone]

This is OPS of a famous hi-fi manufacturer. Their amps're very fast.
Look into the schematic,

Output by L-MOSFET, Exicon ECX10N20/ECX10P20, MJE340/MJE350 as drivers.
Gate stopper resistors have very high value. 470R for P channel and 820R for N channel.
Decoupling for drivers and for output transistors.

If i use low value Gate stopper and adding few caps to G - D pins to compensated. Good value which i can start?! 47pF, 100 or 220pF?

Thanhks!

 

[jacco vermeulen]

GM Telos 2500 ?

 

[Walkalone]

Telos 1000, Telos 2500 similar.
Telos 5000 used KSC2690/KSA1220 as drivers. 72 pcs for 72 output trans

 

[Bob Cordell]

This is OPS of a famous hi-fi manufacturer. Their amps're very fast.
Look into the schematic,

Output by L-MOSFET, Exicon ECX10N20/ECX10P20, MJE340/MJE350 as drivers.
Gate stopper resistors have very high value. 470R for P channel and 820R for N channel.
Decoupling for drivers and for output transistors.

If i use low value Gate stopper and adding few caps to G - D pins to compensated. Good value which i can start?! 47pF, 100 or 220pF?

Thanhks!

This is a very interesting design, in that they used separate driver pairs for each output pair. With so much isolation from the output devices, it is unclear why they did not use a single set of drivers, unless perhaps it was for power dissipation if they are running the drivers really hot.

I can't read the values, but are they running the drivers from boosted rails?

Yes, those stopper values seem high, although 470 is not extremely uncommon in Lateral MOSFET designs. Maybe they increased the size of the stoppers for the NFETs to obtain some form of approximate matching at HF.

Cheers,
Bob