Bias voltage for Class A/B output and output transistor choices

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I'm finalizing my DIY class A/B amp design now and am wondering some theory on keeping distortion low. This is essentially the design I'm basing my design on.

http://mosfetaudio-didik.com/wp-content/uploads/2011/09/Fig-2.png

As the current swings up and down in the second stage that Q4 drives, the drop across the 1k potentiometer fluctuates from the DC bias where it is about 5V to anywhere from 2V to 20V if a large signal comes in. My intuition tells me that this is not good and that you want the bias voltage across the gates of the output mosfets to stay as constant as possible to keep distortion low. What happens to the output as this bias voltage fluctuates? My idea was instead to use a smaller potentiometer with diodes in series, which would have a much more constant voltage drop across them as the voltage fluctuates and still allow for fine tuning of the bias voltage.

Also what should I be looking for when choosing transistors to use? I currently was going to use IRF530 and IRF9530 but these seem pretty outdated and I wasn't sure if theres any difference between output transistors besides voltage and current capabilities. Also the same goes for the driver transistors. I built a low power model of my design with just 2n3904's and 2n3906's for all the diff. pair transistors and a TIP30 for Q4. Is there anything wrong with these?

Thanks for any help!
 
Moderator
Joined 2002
Paid Member
Hi

1. You can use a CCS in place of the bootstrap components. This will maintain a constant current through the bias circuit. Bootstrapped VAS tends to have more H2, which some people prefer sonically.

2. Switching lateral to vertical MOSFETs will require a bias spreader in place of the trimmer to compensate the tempco of the output devices. The spreader can be made with a MOSFET or with a standard BJT, but the BJT tends to overcompensate and hence needs a small portion of the bias to be fixed. Bob Cordell has a nice description and some practical circuits in his book.

3. You can eliminate crossover distortion with as low as 50-60mA on a pair of vertical MOSFETs. However you will have some gm droop issues at such low bias. I have excellent results with 150mA, and I wouldn't go much lower than that. Of course the power supply voltage will determine the dissipation and you may need to size heatsinks accordingly.

4. Vertical MOSFETs tend to have a significant input capacitance. You need to see whether your VAS can adequately drive them (my guess is no) without significant distortion. Bipolar also impose some current load on the driving stage, and this needs to be very carefully considered, right from the VAS stage itself. You are looking for decent current gain, sufficiently high ft and adequate SOAR fro your application.
 
I'm finalizing my DIY class A/B amp design now and am wondering some theory on keeping distortion low. This is essentially the design I'm basing my design on.

http://mosfetaudio-didik.com/wp-content/uploads/2011/09/Fig-2.png

As the current swings up and down in the second stage that Q4 drives, the drop across the 1k potentiometer fluctuates from the DC bias where it is about 5V to anywhere from 2V to 20V if a large signal comes in. My intuition tells me that this is not good and that you want the bias voltage across the gates of the output mosfets to stay as constant as possible to keep distortion low. What happens to the output as this bias voltage fluctuates? My idea was instead to use a smaller potentiometer with diodes in series, which would have a much more constant voltage drop across them as the voltage fluctuates and still allow for fine tuning of the bias voltage.

Also what should I be looking for when choosing transistors to use? I currently was going to use IRF530 and IRF9530 but these seem pretty outdated and I wasn't sure if theres any difference between output transistors besides voltage and current capabilities. Also the same goes for the driver transistors. I built a low power model of my design with just 2n3904's and 2n3906's for all the diff. pair transistors and a TIP30 for Q4. Is there anything wrong with these?

Thanks for any help!

Hi,

This circuit is specifically designed for the lateral MOSFETs (K1058/J162). Their bias conditions and input capacitance allow biasing with a resistor and driving them from VAS directly.

Switching to HexFETs (such as IRF530/9530) will require rather significant changes:
1) You will need a bias spreader - transistor, placed on the main heatsink, tracking the temperature of the output HexFETs and regulating the bias accordingly; otherwise you end up with thermal runaway.
2) You will need an emitter follower in front of the HexFETs - their high non-linear input capacitance requires driving them with low output impedance stage.

As an example, see my design of VHex amplifier. Note the the spreader and the EF buffer in front of the HexFETs.

Attached picture is showing what the boards look like (this is VHex+ with 2 pairs of output devices).

We sell the boards, by the way, if you are interested, $13.90 for the board, $26.90 for the set of 2 boards.

Cheers,
Valery
 

Attachments

  • VHex-plus.jpg
    VHex-plus.jpg
    966.9 KB · Views: 429
Hi,

If your interested in low distortion all I can say is - god what an awful design.
(Its a thermal matching nightmare amongst lots of other things that aren't good.)

You need to read up. Douglas Self is good place to start for low distortion.

rgds, sreten.

Distortion In Power Amplifiers


I'm reading through the article you posted now. Why do you think this is a bad design for low distortion? With a constant current source for the common emitter stage and attaching that to the heatsink to help thermal stability and upping the VAS current to supply enough current to the gates of the mosfets seems like it should fix most of the problems that I see so far. I understand that this may not be a good design and that I may be better off with a different approach but this is for my capstone analog project and I would love to just learn what I'm doing wrong here.

Also thanks to everyone who posted responses. It definitely helped me to understand what's going!
 
Also how would I go about getting an idea of how much current the gates of the mosfets are taking to drive? All I found on the datasheet was input capacitance, but is that in reference to the drain or the source?

Input capacitance:
Ciss = Cgs + Cgd, Cds shorted

Gate-to-drain capacitance, Cgd, is a nonlinear function of voltage and is the most important parameter because it provides a feedback loop between the output and the input of the circuit. Cgd is also called the Miller capacitance because it causes the total dynamic input capacitance to become greater than the sum of the static capacitances.

See the paper on HexFETs here:
https://www.google.ru/url?sa=t&sour...gkdYpSWXUP7ygfJWA&sig2=EuZ8oklJfqsZjejZNiAZsg

Lateral FETs have got smaller and less non-linear Ciss. Also, as I mentioned, the bias is lower and thermal runaway is not a problem.
 
Last edited:
I would add a better VAS circuit with CCS.
Then add proper Vbe multiplier.

I will probably cause outrage saying this but I have got rid of crossover distortion with as little as 10mA bias current. Peavey believe in small bias currents too.

I have designed a few vertical mosfet amps and found I had no bias method.
So I input sine wave and monitor output.
I turn up bias current until crossover distortion goes.
 
I would add a better VAS circuit with CCS.
Then add proper Vbe multiplier.

I will probably cause outrage saying this but I have got rid of crossover distortion with as little as 10mA bias current. Peavey believe in small bias currents too.

I have designed a few vertical mosfet amps and found I had no bias method.
So I input sine wave and monitor output.
I turn up bias current until crossover distortion goes.

That's ok - you get rid from the crossover, visible on the oscilloscope at 10mA. But then, if you measure THD with high precision, you will see noticeable decrease at 80-100mA. This is where the optimum for class A/B is - further quiescent current increase will not give significant improvement, but increase the heat dissipation. Talking about the HexFETs now.

For laterals, the optimum is 120-150mA per output pair.
 
Moderator
Joined 2002
Paid Member
1. For input drive current requirements, use the equation relating slew rate and capacitance. Usually a slew rate of 10-12V/uS is enough.

2. Function of the speedup capacitor is to bypass the bias circuit at high frequencies. The DC conditions remain unchanged. There is some contention about the value of this capacitor, with implementations ranging form 10pF to 220uF. I find about 47uF a decent value for almost all applications. There are some other techniques that can be used to perform the same function, such as bypassing the top resistor of the Vbe multiplier. This requires a much smaller capacitor, and the HF response of the bias circuit is greatly improved.
 
Didn't see any comment about the 0r22 drain resistors in the output stage.

I wonder if these were meant to go in the Source lead where they are normally seen?

No I don't believe so. In a bjt they are necessary to be able to bias the current correctly, but with a mosfet since the vgs to current relationship is less sensitive than the vbe to current relationship of the bjt and therefore while definitely harder to bias without the source resistors, it's very doable with mosfets. And by removing the source resistance you lower your output impedance as well. They are still necessary on the drain though to be able to measure the idle current to bias correctly or else you would have no way to measure the current. At least this is my understanding.

Also one other side note is, the 100 ohm resistors which effectively lower the rails for the first two stages compared to the output stage, why is that necessary? I've heard it's so the drivers clip before the output but why is that better?
 
That's ok - you get rid from the crossover, visible on the oscilloscope at 10mA. But then, if you measure THD with high precision, you will see noticeable decrease at 80-100mA. This is where the optimum for class A/B is - further quiescent current increase will not give significant improvement, but increase the heat dissipation. Talking about the HexFETs now.

For laterals, the optimum is 120-150mA per output pair.

This is a more subjective opinion as it hasn't been well tested, but I believe that below .1% distortion, you'd be hard pressed to notice well any kind of distortion. .1% means the distortion is -60db. At about 10k where at least my hearing is most sensitive, if I played music and then added a 10k signal at -60db I'd be hard pressed to hear it and I plan to do that experiment as well. If a 10k sine wave can't be heard at -60db then I doubt any harmonic distortion could be heard at -60db. Intermodulation distortion is another beast though.
 
No I don't believe so. In a bjt they are necessary to be able to bias the current correctly, but with a mosfet since the vgs to current relationship is less sensitive than the vbe to current relationship of the bjt and therefore while definitely harder to bias without the source resistors, it's very doable with mosfets. And by removing the source resistance you lower your output impedance as well. They are still necessary on the drain though to be able to measure the idle current to bias correctly or else you would have no way to measure the current. At least this is my understanding.

Also one other side note is, the 100 ohm resistors which effectively lower the rails for the first two stages compared to the output stage, why is that necessary? I've heard it's so the drivers clip before the output but why is that better?

AndrewT is right - resistors at the drains don't really make sense.

For the lateral FETs you can go with no resistors at all, even with paralleled output devices. For HexFETs source resistors are a must, even with one pair (same as with BJTs).
 
This is a more subjective opinion as it hasn't been well tested, but I believe that below .1% distortion, you'd be hard pressed to notice well any kind of distortion. .1% means the distortion is -60db. At about 10k where at least my hearing is most sensitive, if I played music and then added a 10k signal at -60db I'd be hard pressed to hear it and I plan to do that experiment as well. If a 10k sine wave can't be heard at -60db then I doubt any harmonic distortion could be heard at -60db. Intermodulation distortion is another beast though.

I'm normally dealing with distortion levels in the range 0.0005%-0.005%. Music is not a sine wave - harmonics sum up, becoming much more audible, than you can hear adding a single sine wave component. Different distortion profiles add different flavors to the overall sounding. Crossover distortion is very bad one - it's wide bandwidth, harmonics-rich, very audible. "Harsh" one.

Intermodulation also depends on harmonic distortion. For example, it's virtually impossible to have 14+15KHz intermodulation distortion lower, than 1KHz THD. I mean... my experience is very practical - I did thousands of measurements.
 
Ok I see what source resistors are necessary for load sharing if you have multiple pairs of HexFETs, but I don't think it would be necessary for a single pair.

Also I agree that music is not just sine waves, but why then measure with a 14 + 15k IMD measurement and not a multi-signal IMD measurement? It seems to me that that would approximate a real signal most accurately. Or even better yet a completely complex input like that of a few instruments, and then measure the distortion at the output? It doesn't seem to me that that would be much harder. And if that isn't a reality then it doesn't seem reasonable to me that having .01% IMD would lead to any more than .1% IMD with music even with the added complexity. Furthermore this is all being measured at maximum rated power at which point any real loudspeaker would be completely beyond any of these distortion measurements.

Sorry didn't mean to get distracted from the topic, but I find it interesting that we all talk about harmonic distortion or IMD distortion and how those would translate into the distortions of music, but why then don't we just input a musical signal into an amplifier, subtract the input signal from the output and see what distortion there is?

Back to the topic at hand though, one other question I had was for lateral mosfets, if you had no source or drain resistors, how would you measure the idle current?
 
Ok I see what source resistors are necessary for load sharing if you have multiple pairs of HexFETs, but I don't think it would be necessary for a single pair.

Also I agree that music is not just sine waves, but why then measure with a 14 + 15k IMD measurement and not a multi-signal IMD measurement? It seems to me that that would approximate a real signal most accurately. Or even better yet a completely complex input like that of a few instruments, and then measure the distortion at the output? It doesn't seem to me that that would be much harder. And if that isn't a reality then it doesn't seem reasonable to me that having .01% IMD would lead to any more than .1% IMD with music even with the added complexity. Furthermore this is all being measured at maximum rated power at which point any real loudspeaker would be completely beyond any of these distortion measurements.

Sorry didn't mean to get distracted from the topic, but I find it interesting that we all talk about harmonic distortion or IMD distortion and how those would translate into the distortions of music, but why then don't we just input a musical signal into an amplifier, subtract the input signal from the output and see what distortion there is?

Back to the topic at hand though, one other question I had was for lateral mosfets, if you had no source or drain resistors, how would you measure the idle current?

This needs to be done with caution, very carefully, having reliable connections - just connecting an ampere-meter in series with one of the drains. Or in series with one of the rails, subtracting the current, consumed by the front-end section.
 
This needs to be done with caution, very carefully, having reliable connections - just connecting an ampere-meter in series with one of the drains. Or in series with one of the rails, subtracting the current, consumed by the front-end section.

That makes sense, but why not just use a small drain resistor? It would make the biasing a hell of a lot easier and with a small enough resistor wouldnt lead to too much wasted heat/headroom
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.