TGM7 - an amplifier based on Greg Ball SKA

With the public release of the schematics for Greg's very well regarded power amplifier "Simple Killer Amp" !... I am going to start a new build.

I have dubbed this the 'TGM7', it's how I number my projects. This is the 7th one. Each project I try to do something different to learn something new. The new elements for this project that interest me are i) MOSFET outputs (never used them before), ii) double LTP input (never used this before), iii) Greg's novel topology for high PSRR and iv) I want to use a commercial pcb vendor instead of making my own pcb. I plan to use Freeware Eagle to design the pcb and I'd welcome help.


What I plan to do differently from Greg's original GB150 ?
- since Greg is not releasing his pcb layout I'd like to develop a new one, possibly using a different form factor (Eagle Freeware is limited to 100mm x 80mm anyway).
- component choices will be looked at, some of the original devices are not easily found. I am also not shy to use surface mount parts as there are many more choices of good components in this format. Many surface mount parts are relatively large and do not need to be a chore to use. It allows for a compact pcb design with less solder and less risk of bad joints.

If anyone else is interested feel free to weigh in with your advice etc.


EDIT: some things changed after I started this thread so I have edited this opening post:

1) I'm building with two pairs of FETs, same as Greg
2) Modified current sources.

The final version of this amplifier works very nicely and has seen a lot of use in my house.

post #8 for initial schematic
post #160 for photo of final pcb
post #183 for as-built schematic
Post #195: zip file for pcb if you want to make yourself


Oh, and by the way, Paulo built his own version with through-hole parts :)
 
Last edited:
With Eagle lite, you can draw outside that limit and include it within your PCB boarder, you just cannot place parts from the schematic page there. SMDs are the new way to go, just be aware of Pd derating. I use tiny SMD extensively without issue.

Q-fet is Fairchild's designation for the planer stripe type vertical mosfets. This style of fet is produced by other companies now but I think Fairchild's Q-fets were some of the first ones produced. Minor differences compared to cellular hex type devices is a sharper increase in Cgd at Vds saturation, and a greater dependence of Gm upon Vds for a given Vgs. This may or may not be an issue depending on the circuit. They do seem to be able to take more abuse though. I like them quite a bit but I have only used them in class A or with some sort of local error correction such as HEC. I have noticed that the only thing different between FQA designation and FQP is the package size. Some of them are available in both packages and the only difference in the datasheet is Pd and Pd derating. They are the same die. Also, I have found that if you can find a way to eliminate the mica under the tab and mount them directly to a larger hunk of metal like in this amp, the TO-220 devices will surprise you at the amount of power they can handle.
 

Attachments

  • DSCF3786.JPG
    DSCF3786.JPG
    834 KB · Views: 4,449
  • DSCF2325.jpg
    DSCF2325.jpg
    691.7 KB · Views: 4,147
I hate SMDs... :crying:

Cheer up Paulo, it really isn't that bad, honestly. Besides, not all the parts would be SMD, it will depend on what is the best choice and we haven't starting to look at the parts yet.

A sugestion: design a cd servo instead of the NFB cap. DC coupled the amp sounds fabulous!

It sounds exciting but I think it's for a phase 2 project, first build should be close to the original design. Let's choose some nice caps for the feedback so as not to have to worry about them.
 
Hi Hugh,

Well it's slow - on accounts of nice weather, a good movie and the local open air Jazz festival (followed by a business trip next week).

Anyhow, I went ahead and plugged the baseline schematic into Eagle. This of course means choosing part sizes. I decided to stick with through-hole parts - partly as I like the idea of reproducing the essence of the original. It's fun - reminds me of when I started out by cloning AKSA (it's still playing).

I may also keep room for two pairs of output devices if it doesn't get too crowded, but the intent for my build is to use just the one pair and with +/-50V rails still get full power. I did find reference somewhere to a GB75 pcb design that Greg was working on some time ago with a single pair but it seems he never finished it.

I see a couple of tweaks that will have to be incorporated yet - the first being to split the resistor in the tails of the current source/sink and move the end of rail cap to the junction between these resistors - a classic and popular approach to improving the PSRR of the current source for nearly zero cost and effort.

The other change I'm mulling over is to take out the temperature sensing diode-wired-transistors. It seems to me that the master device in the current source/sink can be used as the temperature sensing element all by itself (simulations seem to show it works) with some emitter degeneration to tune the temperature sensitivity - simplifies the circuit a little, but I wonder if there is some nasty consequence I'm overlooking ?

Any tweaks and mods you'd suggest ?
 

Attachments

  • sch.jpg
    sch.jpg
    104.1 KB · Views: 4,366
Last edited:

fab

Member
Joined 2004
Paid Member
Hi Bigun

I will be following this thread closely since I do intend to listen to an SKA GB150 amp:p

Regarding the output mosfet I am interested in suggestions as long as the VGSoff is high enough for the SKA amp to work properly at full power.

As for mods I would put separate source resistor for each mosfet allowing less need for VGS close match...

If you are able to take out the temperature sensing diode-wired-transistors I would try it also in my CFB version. However, if Greg Ball has inlcuded it I suppose that it was to obtain the proper compensation for the mosfet VGS drift where VBE drift is different...

Fab
 
Last edited:
I had a look at the FQP12P20 QFET and it has -ve Vgs temperature coefficient all the way up to 10A. This means that there is very real risk of hot spots on the die. Modern devices push this temperature coefficient threshold higher and higher and have made SOA a headache in linear use. Tempco at the 100mA mark is about 5~6mV/C
 
Regarding the output mosfet I am interested in suggestions as long as the VGSoff is high enough for the SKA amp to work properly at full power. As for mods I would put separate source resistor for each mosfet allowing less need for VGS close match...

As far as I can see, most of the vertical MOSFETs we might consider using here will have enough Vgs room to work in SKA and that's the case for the Fairchild devices I'm considering. There are Laterals that have quite low Vgs and for these Greg designed the GB100 - although I don't think it ever came to market. Good idea on the separate source resistors, I wonder why Greg didn't do this, must be a reason. However, I think it's still sensible to match the devices well as the source resistor value used is not exactly generous.

Edit: I'm suspect Greg added this resistor to reduce turn-on current when the devices are cold.

If you are able to take out the temperature sensing diode-wired-transistors I would try it also in my CFB version.

I didn't explain well - I will still incorporate transistor temperature sensing elements as they are needed. The thing is, if you look at the schematic I posted above there are three transistors in the current source / sink. They are all temperature sensitive. If you remove the diode-wired transistor you are left with the classic 2-transistor source/sink. In the 2 transistor version you can still place one of the devices (the master device, the one at most left of the diagram, Q6 and Q8) onto the heatsink and as it warms up it will reduce the current level of the source/sink. So I wonder why not do this ?


I had a look at the FQP12P20 QFET and it has -ve Vgs temperature coefficient all the way up to 10A.

Thanks for posting that paper, most relevant. I did look at the datasheet for the FQA40N25 on Digikey's website. Sure enough if you look at the Vgs verses Id curves at around the 100mA level you need to swing the gate 1V to keep current stable between 25degC and 175degC - around 12.5mV per degC. That's a lot but only twice that of the devices Greg used. Well this is interesting, part of the fun of learning about using FETs instead of BJTs.

I don't think any "conventional package" mosfets would allow a single pair to survive 50V rails into real loads

These QFET parts have about twice the SOA of the parts Greg used, why wouldn't one of them do the job of two ? - it's not for welding duties, just music ??
 
Last edited:
Properly biasing vertical mosfets in common source is hard.
You need 6-10mV/C for ambient compensation and anything up to double that for die temperature rise ABOVE ambient, depending on how well the sensor is tracking the die. GB has done a good job at compensating for ambient, set by the ratio of the 3k and 1k1 resistors times the sense diode tempco, so about 6mV/C. He may well be undercompensating for die temperature rise, depending on heatsink size. This is probably why the SKA has a bit of a reputation for blowing with square waves etc.
I am thinking about using a simple PIC processor to sense ambient and die delta separately and do the maths properly
 
I will take a look at what I can do to ramp up the compensating tempco so that there is the flexibility to use different MOSFETs.

I didn't hear about square wave issues with the amp - but I'm a bit skeptical about square wave testing myself - I've read more than one well-reasoned report why this type of testing can be misleading. And why a 20KHz high power square wave will fry an output stage faster than you can set the gain on the scope... !
 
Yes, David, well spotted.

This devices are around -12mV/C tempco on the Vgs according to measurements I made a couple of years on the FQA series, which are rated to 40A and 300W.

You need a fairly hefty bias generator to accommodate this soaring rise of current with temperature.

Cheers,

Hugh

I found the tempco of these mosfets to be slightly worse than the plural well devices as well. However, at least with a HEC scheme, adequate compensation can be had by using both error amp devices monitoring the thermal difference for larger Vgs muliplication, I seem to recall Bob used one of the devices as the Vgs multiplier in his EC hexfet amp. HEC is source follower (common drain) and I have not used these devices for class AB common source. I suspect source ballast resistors would help to reduce the tempco variation but it may be a challenge to accurately track the tempco. Fortunately mosfets have a larger range of acceptable bias than do BJTs related to the larger variation in Gm. I suggest starting with a slightly higher bias and overcompensating just a bit. You will have to build an actual real circuit :eek: and do some trial and error.:) Don’t be too misled by the Pd max figure, Pd de-rating is significant for these devices. For example, FQP47P06 Pd max is 160W @ 25C, but de-rated at 1.06 W/C. So about 27W at 150C, still quite significant for a TO-220 device.:cool: Another small advantage of the planer stripe fets is a lower gate charge vs drain current di/dt. As far as hot-spotting, the type of mosfet is important. Certain types are able to more evenly spread the heat across the die. I have found the planer stripe, at least the Q-fet brand, to be surprisingly quite robust when used as an analog power device. (notice I did not say linear, there is nothing linear about these devices:D)



And why a 20KHz high power square wave will fry an output stage faster than you can set the gain on the scope... !

Not always true. I have made amps such as this one that can output much faster square wave than 20KHz and it did not fry.:)
 
Last edited:
Attached shows any alternative current source topology. Don't be distracted by the part numbering which is different compared with the master schematic.

PSRR: for improved PSRR you see the small rail capacitor is connected not to the collector of the current source 'master device' Q21, but to the mid-point between the two tail resistors. I've played around with the values a bit and it's possible to tweak up the PSRR in simulations but it can get sensitive to local bias conditions and I would not rely on it to achieve the highest possible PSRR - nevertheless it is an improvement over the baseline published by Greg recently. And if you read around on the internet you'll find that others have tinkered with the baseline current source design, including the use of a zener diode to define a 'regulated' voltage at this junction between the two resistors.

Tempco: In this version of the current source I'm exploring the option to remove the diode-wired transistor and instead I am using the current source 'master transistor' itself as the temperature sensing element which should be placed in physical contact with the output MOSFETs or related heatsink. When this device warms up the output of the current source is reduced and hence so is the bias to the output MOSFETs - which is how we want it to operate. The tempco sensitivity of this arrangement can be adjusted, it depends on the emitter degeneration of Q21, i.e. R93 in this diagram.

In my simulation I varied the temperature of Q21 between 30 degC and 50 degC for a total delta of 20 degC. I then 'probed' the simulated voltage at the gate of the upper output MOSFET to see how much it changed.

With the values shown, the simulated voltage change is 135mV, which is 6.75mV per deg C. This is not far from the tempco of the MOSFETs recommended by Greg.

The largest tempco is achieved when the emitter degeneration resistor on Q21 is set to zero. The value of the 'sense resistor' R49 is adjusted to achieve the desired idle current through the output MOSFETs. Simulating the circuit using these values predicts a tempco of 16mV per deg C. Could this be enough for safe use of the QFETs ?
 

Attachments

  • currentsource.jpg
    currentsource.jpg
    8.8 KB · Views: 1,719
Last edited:

fab

Member
Joined 2004
Paid Member
Attached shows any alternative current source topology. Don't be distracted by the part numbering which is different compared with the master schematic.
.....

Tempco: In this version of the current source I'm exploring the option to remove the diode-wired transistor and instead I am using the current source 'master transistor' itself as the temperature sensing element which should be placed in physical contact with the output MOSFETs or related heatsink. When this device warms up the output of the current source is reduced and hence so is the bias to the output MOSFETs - which is how we want it to operate. The tempco sensitivity of this arrangement can be adjusted, it depends on the emitter degeneration of Q21, i.e. R93 in this diagram.

In my simulation I varied the temperature of Q21 between 30 degC and 50 degC for a total delta of 20 degC. I then 'probed' the simulated voltage at the gate of the upper output MOSFET to see how much it changed.

With the values shown, the simulated voltage change is 135mV, which is 6.75mV per deg C. This is not far from the tempco of the MOSFETs recommended by Greg.

The largest tempco is achieved when the emitter degeneration resistor on Q21 is set to zero. The value of the 'sense resistor' R49 is adjusted to achieve the desired idle current through the output MOSFETs. Simulating the circuit using these values predicts a tempco of 16mV per deg C. Could this be enough for safe use of the QFETs ?

Hi Bigun

This is interesting but quite unusual current source with R93 instead of emitter of Q21 connected to power rail.... How do you predict by calculation the voltage across R49?

Thanks
Fab