Charles, I am a 'try anything' kind of guy. Just because YOU didn't like it, doesn't mean that it can't be useful to me. I mean this in the best sense. Besides, I like to be polite.🙄
Well, I have to admit that I never tried it in an actual amplifier. The concept seemed very cool (even though it does involve feedback). But when I did simulations, it never seemed to do much for me. I got more benefit (in terms of linearity) simply by adding more output devices. This was in the context of a bipolar output stage (triple emitter follower or "T circuit"). YMMV.
PS -- Not too many people have every accused me of being polite. And I'm not so sure about you either! (joke)
PS -- Not too many people have every accused me of being polite. And I'm not so sure about you either! (joke)
john curl said:
Apart from a vertical mosfet instead of a BJT ?
What is fascinating is that you came up with the JC-3 topology such an audiolength ago, one which Mr Hansen admits to have "confiscated"
Yet, with the aid of advanced components it can evolve over time into the contemporary JC-1.
The bias arrangement of the JC-1 is the same, even the input section basically is, including the resistor value that connects top and bottom JFETs.
The personal signature of you "Fella Designers" is intrigueing, makes me anxious about the next JC-X power amp.
Why you continue to put up with the feedback parade thread is lightyears beyond me though.
(i feel right at home among the non-polite, no joke)
Descartes? I am sorry to have been sent over to the feedback thread. It is truly a waste of time.
Makes two of us, i would love to eat at Chez Panisse some day.
(ps: maybe some day, some diya folks are willing to have their shoe for dinner there)
(ps: maybe some day, some diya folks are willing to have their shoe for dinner there)
It is possible to have 'lunch' there with a reservation a few days in advance. Impossible for dinner, I am told. Months in advance, but the food is FRESH and good!
Thermal Stability of BJTs
As mentioned in some earlier posts, the value of the chosen emitter resistor in a Class-AB output stage is intimately tied in with output stage thermal bias stability. It is also tied in with what the optimum Class-AB bias should be.
We can use the thermal bias stability equation that I discussed in some earlier MOSFET-related posts to take a closer look at this for BJT output stages. That equation looked at the short-term local thermal stability of an individual output transistor, independent of the heatsink and assuming that the heatsink’s thermal inertia prevented its temperature from changing over the short time intervals being considered. In other words, this aspect of thermal stability can be a problem no matter how big the heatsink is. The thermal runaway process can occur before the heatsink temperature has time to react. It is a problem that is not addressed by the usual thermal-tracking Vbe multiplier for this reason.
The equation addresses the possibility of thermal runaway of the device under consideration, either with the opposite-sex devices, or in terms of current hogging against the same sex devices in parallel. The equation also assumes that the output node is held firmly in place, voltage-wise, by the plurality of other devices in parallel with the device under consideration. To the extent that this assumption does not hold (i.e., some output node voltage compliance) the problem will be mitigated a bit.
Let’s take a look at thermal bias stability for a 150-Watt TO-247 power transistor in a high power amplifier using 0.1 ohm emitter resistors. Assume that this amplifier has rails that can go as high as 90-Volts and that it uses a multiplicity of output transistor pairs. If the output transistor were ideal, we would bias it for 26 mV across its RE for a bias current of 260 mA. As mentioned in earlier posts, the real-world limitations of the transistor (base resistance, beta and emitter resistance) tend to make the optimum bias drop with a 0.1 ohm RE closer to 15 mV, for an idle current of 150 mA per transistor.
The formula estimates the resulting thermal positive feedback factor as:
Beta = Theta_JS * Vce * TCvce * gm
Assume that the transistor has 0.8 C/W thermal resistance and that the insulating interface has 0.5 C/W, for a total thermal resistance to heat sink of 1.3 C/W. Transconductance gm is the net transconductance of the transistor plus RE at the idle bias point. To be conservative, we use the ideal gm of the transistor at 150 mA to arrive at a net gm of 3.7 S. TCvbe is just 2.2 mV/C.
Beta = 1.3 * 90 * 0.0022 * 3.7 = 0.95
This is right on the borderline of thermal runaway. The above equation and the numbers plugged in are conservative, and there are mitigating factors such as the gm of the real transistor being a bit less, but this is a big warning.
Now let’s substitute a Sanken 2SC3264 200-Watt device in the two-screw package. The device has a thermal resistance of 0.625 C/W and has double the heat sink contact area, bringing the insulator thermal resistance down to 0.25 C/W, for a total of 0.875 C/W.
We now have:
Beta = 0.875 * 90 * 0.0022 * 3.7 = 0.64
This is safer, but any Beta over about 0.5 should call for caution.
Finally, let’s assume that the real transistor gm is degraded by an effective internal RE of 0.1 ohm. This is the same effect that reduced the optimum bias from 260 mA to 150 mA. For gm we now have about 2.7 S. We now have:
Beta = 0.875 * 90 * 0.0022 * 2.7 = 0.47
We are now in safer, but still uncomfortable territory, but we are depending on a real-world transistor imperfection that moves us a bit away from a conservative view. For example, high-beta samples of these transistors might exhibit less of this mitigating imperfection.
Note that there are also other effects that could push thermal Beta to higher values than predicted here. For example, global thermal mis-tracking that temporarily pushes the overall idle bias to a higher-than-nominal value will increase transconductance and increase Beta. The key thing to bear in mind is that the thermal time constants of the die and the package are relatively short.
This problem of short-term local thermal instability is one that is greatly reduced by the use of ThermalTrak devices, by the way.
Cheers,
Bob
As mentioned in some earlier posts, the value of the chosen emitter resistor in a Class-AB output stage is intimately tied in with output stage thermal bias stability. It is also tied in with what the optimum Class-AB bias should be.
We can use the thermal bias stability equation that I discussed in some earlier MOSFET-related posts to take a closer look at this for BJT output stages. That equation looked at the short-term local thermal stability of an individual output transistor, independent of the heatsink and assuming that the heatsink’s thermal inertia prevented its temperature from changing over the short time intervals being considered. In other words, this aspect of thermal stability can be a problem no matter how big the heatsink is. The thermal runaway process can occur before the heatsink temperature has time to react. It is a problem that is not addressed by the usual thermal-tracking Vbe multiplier for this reason.
The equation addresses the possibility of thermal runaway of the device under consideration, either with the opposite-sex devices, or in terms of current hogging against the same sex devices in parallel. The equation also assumes that the output node is held firmly in place, voltage-wise, by the plurality of other devices in parallel with the device under consideration. To the extent that this assumption does not hold (i.e., some output node voltage compliance) the problem will be mitigated a bit.
Let’s take a look at thermal bias stability for a 150-Watt TO-247 power transistor in a high power amplifier using 0.1 ohm emitter resistors. Assume that this amplifier has rails that can go as high as 90-Volts and that it uses a multiplicity of output transistor pairs. If the output transistor were ideal, we would bias it for 26 mV across its RE for a bias current of 260 mA. As mentioned in earlier posts, the real-world limitations of the transistor (base resistance, beta and emitter resistance) tend to make the optimum bias drop with a 0.1 ohm RE closer to 15 mV, for an idle current of 150 mA per transistor.
The formula estimates the resulting thermal positive feedback factor as:
Beta = Theta_JS * Vce * TCvce * gm
Assume that the transistor has 0.8 C/W thermal resistance and that the insulating interface has 0.5 C/W, for a total thermal resistance to heat sink of 1.3 C/W. Transconductance gm is the net transconductance of the transistor plus RE at the idle bias point. To be conservative, we use the ideal gm of the transistor at 150 mA to arrive at a net gm of 3.7 S. TCvbe is just 2.2 mV/C.
Beta = 1.3 * 90 * 0.0022 * 3.7 = 0.95
This is right on the borderline of thermal runaway. The above equation and the numbers plugged in are conservative, and there are mitigating factors such as the gm of the real transistor being a bit less, but this is a big warning.
Now let’s substitute a Sanken 2SC3264 200-Watt device in the two-screw package. The device has a thermal resistance of 0.625 C/W and has double the heat sink contact area, bringing the insulator thermal resistance down to 0.25 C/W, for a total of 0.875 C/W.
We now have:
Beta = 0.875 * 90 * 0.0022 * 3.7 = 0.64
This is safer, but any Beta over about 0.5 should call for caution.
Finally, let’s assume that the real transistor gm is degraded by an effective internal RE of 0.1 ohm. This is the same effect that reduced the optimum bias from 260 mA to 150 mA. For gm we now have about 2.7 S. We now have:
Beta = 0.875 * 90 * 0.0022 * 2.7 = 0.47
We are now in safer, but still uncomfortable territory, but we are depending on a real-world transistor imperfection that moves us a bit away from a conservative view. For example, high-beta samples of these transistors might exhibit less of this mitigating imperfection.
Note that there are also other effects that could push thermal Beta to higher values than predicted here. For example, global thermal mis-tracking that temporarily pushes the overall idle bias to a higher-than-nominal value will increase transconductance and increase Beta. The key thing to bear in mind is that the thermal time constants of the die and the package are relatively short.
This problem of short-term local thermal instability is one that is greatly reduced by the use of ThermalTrak devices, by the way.
Cheers,
Bob
Hi Bob,
I have seen many cases over the years where bias tracking is not good. There are amplifiers that I increased the emitter resistors in value so that I could regain control over the bias current somewhat. In my view, a slightly less good sounding amplifier is far better than a dead one. Makes a big difference on how a customer perceives you.
The other factor that you had briefly touched on was the heat sink size, or thermal efficiency. Often a circuit may be thermally stable until the cost accountants win their battle with engineering and reduce the size of the heat sink. This is happening more and more often as time goes on.
The thermal track devices are pivotal in their ability to allow accurate bias levels. The accountants will still be responsible for many failed units though. 🙁 You can only reduce the ability to radiate heat so far before things fail.
-Chris
I have seen many cases over the years where bias tracking is not good. There are amplifiers that I increased the emitter resistors in value so that I could regain control over the bias current somewhat. In my view, a slightly less good sounding amplifier is far better than a dead one. Makes a big difference on how a customer perceives you.
The other factor that you had briefly touched on was the heat sink size, or thermal efficiency. Often a circuit may be thermally stable until the cost accountants win their battle with engineering and reduce the size of the heat sink. This is happening more and more often as time goes on.
The thermal track devices are pivotal in their ability to allow accurate bias levels. The accountants will still be responsible for many failed units though. 🙁 You can only reduce the ability to radiate heat so far before things fail.
-Chris
Re: Thermal Stability of BJTs
I just joined this thread recently. Where did you derive this formula?
All of our amplifiers use balanced bridge output stages. So our rail voltages typically run somewhere between 25 and 40 volts. If your formula is accurate, this helps things tremendously.
When we built our latest design using the ThermalTrak parts, we did two real-world tests:
a) Measure bias current and heatsink temperature versus time, starting from a dead cold amp and turning it on. There are several different thermal time constants involved here (triple emitter follower). When we got everything dialed in, the bias would peak at about 45 seconds after turn on. Three hours later when everything fully stabilized (the heatsink is the chassis, and it is 22 pounds of aluminum), the bias was about 5% to 10% lower than the peak. This showed a nice, stable and very slightly negative coefficient.
b) Introduced a thermal step by taking a fully warmed up amplifier, measuring the bias current, and then running at 1/3 power (100 watts) for 10 minutes. Then remove the drive signal and monitor the bias current until it returned to normal. Again, when everything was optimized, we were looking for a slight (roughly 5% to 10%) drop in bias current at high temperature (immediately after the thermal step) that returned to normal relatively quickly (within a few minutes or tens of minutes).
The result is an amplifier that is very reliable, and yet the bias remains very constant under all operating conditions.
I never did such thorough testing on our non-ThermalTrak designs, but we've never had any problem with any of them except for the all-MOSFET design. (As noted in an earlier post, this was tracked down to an output device that had a significant bend in the Id vs Vds curve above 30 volts Vds.) Even then, it was only in a handful of units that were presumably operating at high line voltage and had everything else lined up just wrong.
By the latter, I mean that the Vgs threshold voltage on MOSFET's is all over the map. So the bias voltage must also be all over the map. But we used a Vgs multiplier to develop the bias voltage and also generate the thermal compensation.
The problem is that the Vgs multiplier (or Vbe if you are using BJT's) multiplies the thermal coefficient of the transistor by the same factor that it multiplies the Vgs. So with BJT's the multiplier has a very consistent thermal coefficient. But with MOSFET's the multiplier has a thermal coefficient that varies all over the place as the Vgs threshold of the output devices varies from unit-to-unit. Another reason I don't like vertical MOSFET's. (This isn't a problem with the lateral devices because even though the Vgs threshold varies, it is trivial to run the output devices at their zero tempco point and sidestep the whole problem.)
Bob Cordell said:
I just joined this thread recently. Where did you derive this formula?
All of our amplifiers use balanced bridge output stages. So our rail voltages typically run somewhere between 25 and 40 volts. If your formula is accurate, this helps things tremendously.
When we built our latest design using the ThermalTrak parts, we did two real-world tests:
a) Measure bias current and heatsink temperature versus time, starting from a dead cold amp and turning it on. There are several different thermal time constants involved here (triple emitter follower). When we got everything dialed in, the bias would peak at about 45 seconds after turn on. Three hours later when everything fully stabilized (the heatsink is the chassis, and it is 22 pounds of aluminum), the bias was about 5% to 10% lower than the peak. This showed a nice, stable and very slightly negative coefficient.
b) Introduced a thermal step by taking a fully warmed up amplifier, measuring the bias current, and then running at 1/3 power (100 watts) for 10 minutes. Then remove the drive signal and monitor the bias current until it returned to normal. Again, when everything was optimized, we were looking for a slight (roughly 5% to 10%) drop in bias current at high temperature (immediately after the thermal step) that returned to normal relatively quickly (within a few minutes or tens of minutes).
The result is an amplifier that is very reliable, and yet the bias remains very constant under all operating conditions.
I never did such thorough testing on our non-ThermalTrak designs, but we've never had any problem with any of them except for the all-MOSFET design. (As noted in an earlier post, this was tracked down to an output device that had a significant bend in the Id vs Vds curve above 30 volts Vds.) Even then, it was only in a handful of units that were presumably operating at high line voltage and had everything else lined up just wrong.
By the latter, I mean that the Vgs threshold voltage on MOSFET's is all over the map. So the bias voltage must also be all over the map. But we used a Vgs multiplier to develop the bias voltage and also generate the thermal compensation.
The problem is that the Vgs multiplier (or Vbe if you are using BJT's) multiplies the thermal coefficient of the transistor by the same factor that it multiplies the Vgs. So with BJT's the multiplier has a very consistent thermal coefficient. But with MOSFET's the multiplier has a thermal coefficient that varies all over the place as the Vgs threshold of the output devices varies from unit-to-unit. Another reason I don't like vertical MOSFET's. (This isn't a problem with the lateral devices because even though the Vgs threshold varies, it is trivial to run the output devices at their zero tempco point and sidestep the whole problem.)
Bob's formula is simple enough to understand that it doesn't really need a drawn out explanation, but here goes 😉.
Vbe drops at around 2.2mv/K right? That's the TCVbe in Bob's formula. That temperature coefficient combined with the gm determines the change in current versus temperature change, still following me? Now multiply that by the Vce and you now have the change in power dissipation versus temperature change. Multiply that by the Total thermal resistance to the heatsink and you get the change in temperature versus change in power versus change in temperature (the loop part of the equation).
If a 1K increase produces an increase in power dissipation such that the die temperature increases by more than 1 additional kelvin the local thermal system will enter thermal runaway.
P.S. the Beta term is named as such because this is really a fairly standard D.C. feedback analysis, beta in this case is the gain of a positive feedback loop, if it equals or exceeds 1 the gain of the system becomes infinite and the system is unstable.
Vbe drops at around 2.2mv/K right? That's the TCVbe in Bob's formula. That temperature coefficient combined with the gm determines the change in current versus temperature change, still following me? Now multiply that by the Vce and you now have the change in power dissipation versus temperature change. Multiply that by the Total thermal resistance to the heatsink and you get the change in temperature versus change in power versus change in temperature (the loop part of the equation).
If a 1K increase produces an increase in power dissipation such that the die temperature increases by more than 1 additional kelvin the local thermal system will enter thermal runaway.
P.S. the Beta term is named as such because this is really a fairly standard D.C. feedback analysis, beta in this case is the gain of a positive feedback loop, if it equals or exceeds 1 the gain of the system becomes infinite and the system is unstable.
How would one calculate the emitter (source) resistance needed to make the device(s) thermally stable sans any active compensation between the bases (gates)? I.E. Give a verticle mosfet the same properties as a lateral mosfet. (temperaturewise)Tim__x said:
Sans any compensation you must use the thermal resistance to ambient (global thermal loop instead of local thermal loop). Find the value of gm such that "Theta_JS * Vce * TCvce * gm" is significantly less than one (less than 0.5 would be best), set Re to whatever value is necessary to reduce the gm to the appropriate value.
For a fet, find the the value of the slope of Vgs threshold versus temperature, substitute that for the 2.2mv/K. Finding the the source resistor follows the exact same process as used for bipolars, find needed gm, increase Rs such the gm needs are met.
P.S. This is only a D.C. small signal analysis, and a rather incomplete one at that. Almost all the terms are functions of related variables. Gm varies with current, Vce will vary with current multiplied by the external Re but not the internal, etcetera.
For a fet, find the the value of the slope of Vgs threshold versus temperature, substitute that for the 2.2mv/K. Finding the the source resistor follows the exact same process as used for bipolars, find needed gm, increase Rs such the gm needs are met.
P.S. This is only a D.C. small signal analysis, and a rather incomplete one at that. Almost all the terms are functions of related variables. Gm varies with current, Vce will vary with current multiplied by the external Re but not the internal, etcetera.
anatech said:
Hi Chris,
These ara all good points. While it is certainly good to have big heat sinks, it is understandable that they can be a target of cost-cutting. The FTC standard of being able to run at 1/3 rated power into 8 ohms was fairly conservative, and a lot of manufacturers have abandoned that goal. However, when you take that same amplifier and run it into a 4-ohm load, it finds itself in a much less conservative situation.
I'm not sure what most manufacturers do, but my inclination is to have a thermal circuit breaker on the heatsink that shuts the amplifier down at a temperature no greater than 70C. Beyond 70 c you really get into the danger area of the user being burned fairly quickly if they touch it. As you know, 60 C is the temperature at which you can no longer put your finger on something and hold it there.
From that point of view, sizing the heat sink then becomes a matter of three things:
1) can the output stage and related electronics operate reliably for an indefinite period (say, an hour, with live, loud program material) at a heat sink temperature of 70 C (taking into account transistor deratings for SOA, dissipation, etc.).
2) how long do you want to have the amplifier be able to perform at, say, 1/3 rated power, into an 8 ohm load (or some similar criteria).
3) Lastly, we all know that heat is the enemy of electronic components and their reliability. So in addition to the above, the choice of heatsink size is also goverened by how much long term reliability (life) the manufacturer wants to build into an amplifier.
Cheers,
Bob
Hi Bob,
It has been my experience that most distributors do not properly support the equipment they sell anymore. In fact, they do not want it repaired and see service as direct competition to them. Even their own service depots.
In the late 70's, service depots where well supported by the distributor. We had decent parts markups and dealers prices included a small markup for the depot. We had training seminars and the level of service was much better than it is today. They even stock cosmetic parts for repairing shipping or other damage.
Does anyone have a dead Denon DCD-S10? My faceplate is damaged.
N.L.A.
Today, forget about cosmetic parts for the most part, switches and controls are generally not available in the industry. The early failure is planned actually. You can tell simply by picking an amp up these days and I'm not talking about class D types. Power transformers are also a bit undersized.
-Chris
I shouldn't comment, but I will anyway.
It has been my experience that most distributors do not properly support the equipment they sell anymore. In fact, they do not want it repaired and see service as direct competition to them. Even their own service depots.
In the late 70's, service depots where well supported by the distributor. We had decent parts markups and dealers prices included a small markup for the depot. We had training seminars and the level of service was much better than it is today. They even stock cosmetic parts for repairing shipping or other damage.
Does anyone have a dead Denon DCD-S10? My faceplate is damaged.
N.L.A.Today, forget about cosmetic parts for the most part, switches and controls are generally not available in the industry. The early failure is planned actually. You can tell simply by picking an amp up these days and I'm not talking about class D types. Power transformers are also a bit undersized.
-Chris
Re: Re: Thermal Stability of BJTs
Hi Charles,
What you describe is exactly the right thing to do, in my opinion. You've obviously achieved very good bias stability results, in large part due to the use of the ThermalTrak devices.
What you describe in terms of assessing thermal bias stability with a step is very similar to what I described and tested in my AES paper "A MOSFET Power Amplifier with Error Correction", on my site at www.cordellaudio.com under published papers. At the time, I was showing the difference in thermal bias stability between vertical MOSFET and BJT output stages under like conditions. One of the key things I did was run the amplifier at a fairly high power level and then took away the power and monitored the idle bias current as a function of time. In my tests, the vertical MOSFET amplifier was much more stable than the BJT amplifier.
The formula was derived by me, and is a descendant of the thinking and a formula that I described in that AES paper. After more thought, I convinced myself that the formula I showed in the AES paper was not exactly what I thought best. This formula captures all of the factors affecting thermal bias stability in a very simple expression, and is just based on positive thermal feedback loop gain evaluation. I pretty much showed the derivation in an earlier post, and will try to re-post the relevant part for convenience.
You're right, the direct presence of Vce in the formula shows that at higher voltages one pays a price in thermal bias stability, all else remaining equal. It also shows the importance of using transistors with large heat sink contact area (and good insulators) so as to get the thermal resistance from case to sink down as low as possible. Similarly, it illustrates the value of paralleing output devices.
Cheers,
Bob
Charles Hansen said:
Hi Charles,
What you describe is exactly the right thing to do, in my opinion. You've obviously achieved very good bias stability results, in large part due to the use of the ThermalTrak devices.
What you describe in terms of assessing thermal bias stability with a step is very similar to what I described and tested in my AES paper "A MOSFET Power Amplifier with Error Correction", on my site at www.cordellaudio.com under published papers. At the time, I was showing the difference in thermal bias stability between vertical MOSFET and BJT output stages under like conditions. One of the key things I did was run the amplifier at a fairly high power level and then took away the power and monitored the idle bias current as a function of time. In my tests, the vertical MOSFET amplifier was much more stable than the BJT amplifier.
The formula was derived by me, and is a descendant of the thinking and a formula that I described in that AES paper. After more thought, I convinced myself that the formula I showed in the AES paper was not exactly what I thought best. This formula captures all of the factors affecting thermal bias stability in a very simple expression, and is just based on positive thermal feedback loop gain evaluation. I pretty much showed the derivation in an earlier post, and will try to re-post the relevant part for convenience.
You're right, the direct presence of Vce in the formula shows that at higher voltages one pays a price in thermal bias stability, all else remaining equal. It also shows the importance of using transistors with large heat sink contact area (and good insulators) so as to get the thermal resistance from case to sink down as low as possible. Similarly, it illustrates the value of paralleing output devices.
Cheers,
Bob