Output stage
 

Hi

For what it's worth, the following is merely my humble opinion, so feel free to opine as you will….or not....


When it comes to which device is best suited for driving a reactive load like a speaker or a motor, HEX type fets are the trick. Compared to a bipolar the hexfet will out perform hands down in SOA, speed, and cost. While the BJT may have a higher more linear conductance, it also suffers from secondary breakdown, a major killer of BJTs. A TO-220 hexfet can have a Pd of over 50W at 100 degrees C! I have yet to find a similar package BJT to even come close to that.

The hexfet is a vertical type fet and its conductance is typically more than that of a lateral type, but also it has higher input capacitance. This can be a problem for someone using hexfets in a CFP arrangement; therefore, it is better to use them as source follower, where the excess CISS is effectively bootstrapped. Under the right conditions, they are really hard to beat. Then when you consider the price of these transistors at about a dollar or so each, it really makes you wonder what could be done with them. The question is, what do you have to do to make them work as a low distortion linear amplifier? Truth is, a whole lot.:spin:

First you need to have extra voltage for the gate drive circuit. You need a second, higher voltage low current preferably regulated supply. There is much non-linearity involved with these devices that needs to be corrected for. Check out the static characteristics in figure 1 of the datasheet for FQP50N06 and compare to a lateral fet or BJT. Quite far from an ideal straight horizontal line, and Vgs vs conductance is highly dependent on Vds. I find it slightly amusing that people try to use these devices for linear applications as if they were BJTs or lateral fets.:rolleyes: However, I’m becoming strongly convinced that they really can be used, but they require the help of a few extra parts. The differential EC part of this circuit is very similar to Bob’s EC mosfet output stage, which is based on HEC. In this type of EC, the error amp is designed to have unity gain, but lots of bandwidth. Being able to use fast transistors for this stage is important. The problem is hexfets have a thermal response similar to BJTs in that they need thermal compensation even though hexfets are much more forgiving than BJTs in this respect. Small signal transistors would be more ideal for the HEC but are physically difficult to mount with the OPTs to track temperature. Besides, they would do a poor job of really monitoring the instantaneous temperature of the die, for the perfectionist who wishes for more control.:D Also layout is critical, and mounting the EC transistors to the OPTs complicates this aspect.

What if you could use the common mode bias to control the bias setting? Then you could set a minimum bias so that neither transistor ever turns off. That is another issue with these devices. They don’t handle turn off/turn on very well without creating lots of ‘issues’. This is true for all transistors, but these are especially bad. I have come up with a little circuit that might just have some potential. At least the first proto-type circuit seems to be tracking OK. This is not a circuit that would be easily simulated. I set bias to 200mA and it did not vary more than 10-20mA as the transistor case approached fingertip burning temperature (very undersized heatsink:tongue: ). Adding and removing the load, the bias goes straight to the set point. I still have a lot of fiddling yet to do, but so far it seems to be working good enough to warrant more investigation. As the transistors conduct, there will be a differential mode error that is controlled with the HEC. There is also a common mode error if one wants to maintain constant minimum conductance in the “off” device. A representation of the current through each OPT is conveyed to reference GND (0V). The absolute value of the minimum conductance is used to control the bias spread using a fb loop with a IR300 opto coupler from Vishay. The LED and 2 photo diodes are part of the IR300.

The sine wave is the output (30KHz) and the other is the common mode error signal measured as the difference between points A and B (200mV/div). It seems the bias spread is increased during the peak conduction of each half cycle. I guess this makes sense because of the extra Vgs needed. The wave forms are different for the N-ch vs the P-ch devices. The FQP47P06 seems to need increasingly more Vgs than the N type. Interestingly enough, this is sort of reflected in the data sheets fig 1 for the region of interest.

One problem with the temperature controlled version is that there needs to be a coupling capacitor between the gates of the two hexfets in order to prevent cross conduction. This was a frustrating problem in the last model with the load attached or not. The higher the frequency, the more un-depleted charge builds up on the gates and prevented them from turning off, even within the audio band. However, with active discharge this circuit does not require that cap and does not begin to cross conduct until above 80-150KHz, depending on amplitude. There is indeed a CM error signal that contributes to this behavior.:) But it seems limited in BW to fit in all the HF component of the error signal at such high driving frequencies.:( A point of future investigation….

It turns out there is a lot of funky stuff you have to do to the gate drive of a hexfet in order to get it to work optimally in a linear application, but IMO is well worth it if you can pull it off.