On Semi ThermalTrak

[stinius]

Hi Bob

I like the biasing circuit and the way you use the TT diodes and how you solve the driver and pre-driver temp-tracking problem.

It’s a simplified schematic so I have some questions:

Why do you refer E1 to ground?
Could you please explain the E1-circuit and the implementation of R10?
Is V3 a cascode?

Cheers
Stinius

 

[Bob Cordell]

Hi Bob

I like the biasing circuit and the way you use the TT diodes and how you solve the driver and pre-driver temp-tracking problem.

It’s a simplified schematic so I have some questions:

Why do you refer E1 to ground?
Could you please explain the E1-circuit and the implementation of R10?
Is V3 a cascode?

Cheers
Stinius

Hi Stinius,

Thanks for your kind words. E1 and V3 are there only for purposes of simulation. I apologize for any confusion. R10 and the shunt capacitor just forms an arbitrary low pass filter for stability of the feedback loop I have formed for purposes of simulation. V3 just provides an offset on the order of what the bottom node of the spreader wants to be at, so that the nominal output of E1 is near zero. I'm basically just faking out a VAS.

Cheers,
Bob

 

[stinius]

Hi Stinius,

Thanks for your kind words. E1 and V3 are there only for purposes of simulation. I apologize for any confusion. R10 and the shunt capacitor just forms an arbitrary low pass filter for stability of the feedback loop I have formed for purposes of simulation. V3 just provides an offset on the order of what the bottom node of the spreader wants to be at, so that the nominal output of E1 is near zero. I'm basically just faking out a VAS.

Cheers,
Bob

Hi Bob

Thank you for the clarifying, as you see I was a bit confused.

I’m of the old school and like to use paper and pencil and try to understand what’s happening.
I’m using Multisim, but only to verify the paperwork.

Cheers
Stinius

 

[JPV]

Re: ThermalTrak Bias Spreader continued

If the driver transistors are not mounted on the main heat sink, a useful arrangement that is well-suited to this bias spreader architecture is to mount the pre-drivers, drivers and Vbe multiplier transistor on a single, nearly isothermal, metal bar. The bar also acts as a bit of a heat sink for these components. If all of these devices are at the same temperature, it can be seen that the Vbe multiplier will do a pretty good job of temperature compensating these devices, while the tracking diodes can do a good job tracking the output device temperatures.

As mentioned above, R3 acts to enhance and control the sensitivity to the TC of the tracking diodes. The numbers below show the sensitivity to the tracking diodes as a function of R3. In all cases the spread was adjusted to produce 26 mV across the output emitter resistors. Sensitivity is defined as the change in voltage base-to-base of the output transistors divided by the change in voltage of D1 + D2. Notice that sensitivity is less than unity in a simple arrangement without R3. Also note that in some previous postings there has been some discussion about the need for the sensitivity to be substantially greater than unity. The numbers shown here provide sensitivity values up to about 1.5:1.


R3...... R2...... sensitivity
Infin... 716..... 0.92
20k..... 794... 1.01
10k..... 893... 1.10
7.5k ... 974 ... 1.16
5.0k.... 1190... 1.28
4.0k... 1430... 1.37
3.0k... 2142... 1.52

The current through R3 acts to increase the bias spread by forcing more current through R2. The current through R3 is strongly influenced by the voltage drop across the tracking diodes. As their voltage decreases with increases in temperature, the current through R3 decreases, decreasing the spreader voltage as desired. This is how the sensitivity of the tracking diodes is enhanced. The presence of R3, of course, requires that R2 be made larger for a given output stage bias current.

Because R3 requires that R2 be made larger, the multiplication factor of the Vbe multiplier is reduced somewhat. Ideally, its factor would be 4:1, since the single Vbe of the multiplier is being called upon to compensate for the four pre-driver and driver Vbe’s. The price paid for increased sensitivity to the tracking diodes is some decrease in sensitivity to the multiplier Vbe. With R3 set for 3k and tracking diode sensitivity at 1.52, sensitivity to the multiplier Vbe has decreased to about 2.57.

This means that the pre-drivers and drivers will be a bit under-compensated, but this is a lot better than not being compensated at all as is often the case in conventional arrangements. If the drivers are mounted closely on either side of the multiplier transistor at the center of the bar, and the pre-drivers are mounted further out on the wings of the metal heat spreader bar, some thermal attenuation experienced by the pre-drivers will improve the degree of compensation. The placement of one or two board-mounted diodes in series with R3 would also mitigate the reduction in sensitivity to the multiplier Vbe. Other similar circuit approaches are also possible.

Cheers,
Bob

Bob,


1° First comment:

I can be wrong but this is my analysis. If R3 is infinite the sensitivity should be 1 because Q1 is now biased by a current source in an independent way from D1 and D2 variation. Therefore variation in Vd1 and Vd2 will be reflected totally on the bases of the output transistors.

Quick calculation
let I1 = current in R1, I2 current in R2 and I3 current in R3.

Then if we neglect as usual base current, I1=I2 + I3
I2 = Vbe/R2 and I3 = (Vd1 + Vd2 +Vbe)/R3

Replacing and rearranging we get ( if I am not wrong) with Vd=Vd1+Vd2

Vbe (R1/R2 + R1/R3 +1) + 2Vd (R1/R3 +1) = Vbias

The sensitivity S is the partial derivative of Vbias with respect to 2Vd

S = R1/R3 +1 and if R3 infinite the S=1.


2° A Vbe multiplier is used as I understand to have a multiplication factor and if used as T sensor, to have a drift correction.
In this topology, we can assume that the drift of the pre driver is negligeable. Then if we use thermaltrack transistors as driver also, we can put four diodes is serie tracking and biasing the transistors and a simple potentiometer to compensate for the predrivers Vbe and to adjust for output bias. This of course if we have a current source in the Vas.
If we do this, is it then not good to bypass the Vbias with a good polyprop capacitor close to the Vas "pins" ?

What do you think?

JPV

 

[Bob Cordell]

Re: Re: ThermalTrak Bias Spreader continued

Bob,


1° First comment:

I can be wrong but this is my analysis. If R3 is infinite the sensitivity should be 1 because Q1 is now biased by a current source in an independent way from D1 and D2 variation. Therefore variation in Vd1 and Vd2 will be reflected totally on the bases of the output transistors.

Quick calculation
let I1 = current in R1, I2 current in R2 and I3 current in R3.

Then if we neglect as usual base current, I1=I2 + I3
I2 = Vbe/R2 and I3 = (Vd1 + Vd2 +Vbe)/R3

Replacing and rearranging we get ( if I am not wrong) with Vd=Vd1+Vd2

Vbe (R1/R2 + R1/R3 +1) + 2Vd (R1/R3 +1) = Vbias

The sensitivity S is the partial derivative of Vbias with respect to 2Vd

S = R1/R3 +1 and if R3 infinite the S=1.


2° A Vbe multiplier is used as I understand to have a multiplication factor and if used as T sensor, to have a drift correction.
In this topology, we can assume that the drift of the pre driver is negligeable. Then if we use thermaltrack transistors as driver also, we can put four diodes is serie tracking and biasing the transistors and a simple potentiometer to compensate for the predrivers Vbe and to adjust for output bias. This of course if we have a current source in the Vas.
If we do this, is it then not good to bypass the Vbias with a good polyprop capacitor close to the Vas "pins" ?

What do you think?

JPV

Hi JPV,

You are basically correct on all counts.

If R3 is infinite, the sensitivity to D1 + D2 is theoretically unity. Simulation shows it to be slightly less. This is probably due to a slight bit of bias spread attenuation encountered as the bias spread makes its way to the output transistor bases, as a result of the finite gm of the pre-driver and driver transistors working against their emitter resistors.

BTW, sensitivity was evaluated in simulation by putting a swept dc voltage source in series with the diodes and plotting the voltage from base-to-base of the output transistors. Similarly, sensitivity to the Vbe of the Vbe multiplier transistor was evaluated by putting a swept dc source in series with its emitter and plotting the base-to-base voltage at the pre-driver bases.

You are correct that the temperature change of the pre-driver transistors is usually fairly small, due to their dissipation being fairly small. In the design I illustrated, pre-driver dissipation was about 500 mW.

However, in my preferred implementation described, I chose to deliberately make them largely isothermal with the drivers and Vbe multiplier, so everything could be pretty much temperature-tracked together. Obviously, this is only one of several different ways to use this circuit.

For example, placing the drivers on the heat sink and the pre-drivers by themselves also can be made to work, just with different temeparture-sensing settings. In such an arrangement, the pre-drivers would be essentially exposed to the board ambient. In the absence of ThermakTrak drivers, mounting the drivers on the heatsink might require placement of a Vbe multiplier on the heat sink, introducing thermal lag and extraneous temperature variations into their compensation.

I agree that it is tempting to employ ThermalTrak driver transistors and incorporate their diodes in the compensation scheme. However, I am unaware of any smaller (e.g., TO-220, smaller die) Thermaltrak transistors that would make really good drivers and retain their ft down to the lower operating currents of the drivers and have a bit lower capacitance. I'm just speculating here, and it may be perfectly reasonable to use the full-size transistors for the drivers.

In larger power amplifiers with multiple output pairs and larger driver bias current, the use of ThermalTrak drivers makes more sense. The added SOA in the drivers can be a plus.

Finally, the need for use of ThermalTrak transistors for the drivers is a bit less, since the drivers operate in class-A and are subject to much smaller dissipation and temperature swings.

Cheers,
Bob

 

[JPV]

Re: Re: Re: ThermalTrak Bias Spreader continued

Hi JPV,

You are basically correct on all counts.

If R3 is infinite, the sensitivity to D1 + D2 is theoretically unity. Simulation shows it to be slightly less. This is probably due to a slight bit of bias spread attenuation encountered as the bias spread makes its way to the output transistor bases, as a result of the finite gm of the pre-driver and driver transistors working against their emitter resistors.

BTW, sensitivity was evaluated in simulation by putting a swept dc voltage source in series with the diodes and plotting the voltage from base-to-base of the output transistors. Similarly, sensitivity to the Vbe of the Vbe multiplier transistor was evaluated by putting a swept dc source in series with its emitter and plotting the base-to-base voltage at the pre-driver bases.

You are correct that the temperature change of the pre-driver transistors is usually fairly small, due to their dissipation being fairly small. In the design I illustrated, pre-driver dissipation was about 500 mW.

However, in my preferred implementation described, I chose to deliberately make them largely isothermal with the drivers and Vbe multiplier, so everything could be pretty much temperature-tracked together. Obviously, this is only one of several different ways to use this circuit.

For example, placing the drivers on the heat sink and the pre-drivers by themselves also can be made to work, just with different temeparture-sensing settings. In such an arrangement, the pre-drivers would be essentially exposed to the board ambient. In the absence of ThermakTrak drivers, mounting the drivers on the heatsink might require placement of a Vbe multiplier on the heat sink, introducing thermal lag and extraneous temperature variations into their compensation.

I agree that it is tempting to employ ThermalTrak driver transistors and incorporate their diodes in the compensation scheme. However, I am unaware of any smaller (e.g., TO-220, smaller die) Thermaltrak transistors that would make really good drivers and retain their ft down to the lower operating currents of the drivers and have a bit lower capacitance. I'm just speculating here, and it may be perfectly reasonable to use the full-size transistors for the drivers.

In larger power amplifiers with multiple output pairs and larger driver bias current, the use of ThermalTrak drivers makes more sense. The added SOA in the drivers can be a plus.

Finally, the need for use of ThermalTrak transistors for the drivers is a bit less, since the drivers operate in class-A and are subject to much smaller dissipation and temperature swings.

Cheers,
Bob

From a practical point of view, the use of drivers having the same package as the output transistors allows to mount all of them side by side on the heatsink and press them with a long U aluminium profile across them and screws between each one. By using Belleville washers you can control the pressure and have constant and same pressure on each transistor so the same theta; if Thermaltracks are used in this way in a triple T topology I believe it is a rather optimal design with repect to crossover distortion.

I have finalized a board but I am still waiting for the completion by a friend of the CAD design of my layout.

Two questions:

I followed your great recommendation and bought on ebay a range of Tektronics instruments ( FG504, SG505 and AA501A distortion analyzer), amazing price.
Do you know if breadboarding using 3M breadboards is valuable for these analog applications.

Second, this capacitor found between collector and emitter of Vbe multiplier is there to stabilize the Vbe transistor . If we don't use a Vbe multiplier do you believe that it still make sense to shunt the signal with this capacitor to avoid signal currents flowing in the rather large loop made by the thermaltrack diodes? The loop includes potentiometer of low resistivity, therefore to be efficient the capacitor will have to be large. Is this not a problem?

Thanks for your comments

 

[jackinnj]

Re: Re: Re: Re: ThermalTrak Bias Spreader continued

I followed your great recommendation and bought on ebay a range of Tektronics instruments ( FG504, SG505 and AA501A distortion analyzer), amazing price.
Do you know if breadboarding using 3M breadboards is valuable for these analog applications.

I wonder if I am answering the right question -- there is a guy on EBay who sells breadboards which will fit into Tektronix TM500 and TM5000 series "boxes". They have the fingering which fits into the backplane. The metal cases which house plug-ins for the 5000 series scopes are the same size as the TM5xx PI's -- the scope plug-ins are dirt cheap on EBay.

 

[JPV]

Re: Re: Re: Re: Re: ThermalTrak Bias Spreader continued

I wonder if I am answering the right question -- there is a guy on EBay who sells breadboards which will fit into Tektronix TM500 and TM5000 series "boxes". They have the fingering which fits into the backplane. The metal cases which house plug-ins for the 5000 series scopes are the same size as the TM5xx PI's -- the scope plug-ins are dirt cheap on EBay.

I was refering to breadboarding in general for power amps development. I have seen these boxes to develop a custom module for Tek 5000 range.

Thanks for your comment on general analog breadboarding and development tips for power amps.

JPV

 

[Bob Cordell]

Re: Re: Re: Re: ThermalTrak Bias Spreader continued

From a practical point of view, the use of drivers having the same package as the output transistors allows to mount all of them side by side on the heatsink and press them with a long U aluminium profile across them and screws between each one. By using Belleville washers you can control the pressure and have constant and same pressure on each transistor so the same theta; if Thermaltracks are used in this way in a triple T topology I believe it is a rather optimal design with repect to crossover distortion.

I have finalized a board but I am still waiting for the completion by a friend of the CAD design of my layout.

Two questions:

I followed your great recommendation and bought on ebay a range of Tektronics instruments ( FG504, SG505 and AA501A distortion analyzer), amazing price.
Do you know if breadboarding using 3M breadboards is valuable for these analog applications.

Second, this capacitor found between collector and emitter of Vbe multiplier is there to stabilize the Vbe transistor . If we don't use a Vbe multiplier do you believe that it still make sense to shunt the signal with this capacitor to avoid signal currents flowing in the rather large loop made by the thermaltrack diodes? The loop includes potentiometer of low resistivity, therefore to be efficient the capacitor will have to be large. Is this not a problem?

Thanks for your comments

Hi JPV,

Thanks for your kind words; I have really benefitted from numerous TM-500 modules that I have bought. I had to repair about half of them, but it wasn't extremely difficult, just a little time-consuming. Some of the people on Ebay sell CDs that have many of the technical manuals on them, and this makes repair far easier. In many cases the problem turns out to be something burned out or blown in a power supply regulator, or dirty switch contacts. That you snagged an SG505 and an AA501 is a really great catch!

I use perf board with a 0.1-inch grid all the time for prototyping. I do not use flea clips to mount the components. I just "sew" the components right into the board and connect their leads together on the bottom side. It is amazing how much of the connectivity you can achieve on the bottom side with just the bare leads of the components soldered together. For the rest, I just use insulated wire or bare wire jumpers installed through the top of the board.

I always believe in bypassing the Vbe multiplier. I'll usually bypass the entire Vbe multiplier with a small electrolytic and a 0.1 uF film capacitor in parallel. If the Vbe multiplier transistor is remoted to a mounting on the heat sink, I will usually connect a small 0.01 uF cap directly from collector to base at the transitsor to assure against any possibility of oscillation due to the longer leads back to the circuit board.

Cheers,
Bob

 

[jan.didden]

Hi Bob,

I was just thinking about bypassing the Vbe multiplier in the context of using TT-diodes in it. Using these diodes for very fast (10's of millisecond responses) thermal correction, I would think we would not want to bypass the Vbe multiplier. We would want the Vbe multiplier to track as fast as possible the Vbe changes due to thermal cycling in the output devices. Does that make sense?

Jan Didden

 

[Bob Cordell]

Hi Bob,

I was just thinking about bypassing the Vbe multiplier in the context of using TT-diodes in it. Using these diodes for very fast (10's of millisecond responses) thermal correction, I would think we would not want to bypass the Vbe multiplier. We would want the Vbe multiplier to track as fast as possible the Vbe changes due to thermal cycling in the output devices. Does that make sense?

Jan Didden

Hi Jan,

Given the low impedance of the Vbe multiplier as established by shunt feedback, and the low impedance of the TT diodes (on the order of 2.6 ohms each at a current of 10 mA), I don't think a 10 uF electrolytic across the whole thing will cause a slow-down in response on the oder of tens of milliseconds, but I must admit that I have not looked at this issue closely.

Of course, the TT diode portion may not really need to be bypassed, and the Vbe multiplier portion itself does not need to change really quickly, as most of its action in the arrangement I showed is for temperature compensation of the pre-drivers and drivers.

Cheers,
Bob

 

[stinius]

Hello One and All

I'm sorry that we seem to have reached a bit of an impasse on the use of ThermalTraks, but I think their potential is such that we should press on as best we can. I have had time to do a little more work on it.

Ignoring for the moment the unsolved problem of why the short-term time response of the sense diode does not seem to match up with theory, I have worked out a plan for making the compensation over the long term (100 to 1000 seconds) rather accurate. This just assumes knowledge of the junction-copper thermal resistance and the total copper-heatsink thermal resistance. The scheme is still very much in the realm of theory and SPICE, but it looks promising; with a little tweaking of parameters the bias voltage can be held accurate to a small fraction of a millivolt over the long-term.

I hope to try this for real in the next couple of weeks, and see if the short-term issues can be dealt with empirically.

Hi Doug

It’s good to see that you have some time to work on the TT’s
I think you are on the right track when you look at the long time compensation, let’s just forget the first few seconds.

Bob has made a bias circuit that looks very good; to settle the R2 and R3 factor we need to know the delta T (between junction Transistor and junction Diode) and it will of course also depend on Id.

Bob and Doug:
What do you think about adding a R in series with the Vbe multiplier (inside the Bias circuit) with a Vdrop=2xVRe
In the Bob circuit it would be Vdrop=2x26mV = 52mV, and a VAS=10mA, the R= 5R2.

Maybe I’m too theoretical?

Cheers
Stinius

 

[Bob Cordell]

Hi Stinius,

Theoretically, the 26 mV ideal voltage drop across Re is only that value at room temperature. That number is Vt, which is Proportional To Absolute Temperature (PTAT). In other words, the current that should flow in the output transistor that makes its transconductance equal to 1/Re increases in proportion to absolute temperature.

So although the idea of making that portion of the bias spread responsible for the Re drop a constant value (rather than a value that shrinks with increased temperature) is a step in the right direction, it is still not quite theoretically there.

We may be splitting hairs here, however.

One could also argue that the theoretical desire to have the idle bias increase in proportion to absolute temperature means that the output stage should perhaps be slightly under-compensated; but caution needs to be exercised in going this route.

Cheers,
Bob

 

[jan.didden]

Hi Jan,

Given the low impedance of the Vbe multiplier as established by shunt feedback, and the low impedance of the TT diodes (on the order of 2.6 ohms each at a current of 10 mA), I don't think a 10 uF electrolytic across the whole thing will cause a slow-down in response on the oder of tens of milliseconds, but I must admit that I have not looked at this issue closely.

Of course, the TT diode portion may not really need to be bypassed, and the Vbe multiplier portion itself does not need to change really quickly, as most of its action in the arrangement I showed is for temperature compensation of the pre-drivers and drivers.

Cheers,
Bob

Indeed, it's the dynamic impedance versus that cap. Also, in one version I saw (I think it was your preferred version) the Vbe and the TT's are sort of in series and one could put a cap across the Vbe and not across the TT's. OTOH, given the small impedances, one would have to look at the effect of even 10uF anyway.

Jan Didden

 

[roender]

Hi Stinius,

Theoretically, the 26 mV ideal voltage drop across Re is only that value at room temperature.
Cheers,
Bob

Re = Re (external) + Re (internal).
Re (internal) or parasitic emitter resistance is not know as value, is not even constant, so 26mV is not a value measured over external resistance.

 

[JPV]

Re = Re (external) + Re (internal).
Re (internal) or parasitic emitter resistance is not know as value, is not even constant, so 26mV is not a value measured over external resistance.

Yes and we have to add also Rb/Hfe where Rb is the resistance seen by the base.

JPV

 

[JPV]

Hi Stinius,

Theoretically, the 26 mV ideal voltage drop across Re is only that value at room temperature. That number is Vt, which is Proportional To Absolute Temperature (PTAT). In other words, the current that should flow in the output transistor that makes its transconductance equal to 1/Re increases in proportion to absolute temperature.

So although the idea of making that portion of the bias spread responsible for the Re drop a constant value (rather than a value that shrinks with increased temperature) is a step in the right direction, it is still not quite theoretically there.

We may be splitting hairs here, however.

One could also argue that the theoretical desire to have the idle bias increase in proportion to absolute temperature means that the output stage should perhaps be slightly under-compensated; but caution needs to be exercised in going this route.

Cheers,
Bob

If we have a way ( and we have by changing current in the diode)
to adjust experimentally the tempco of the thermaltrack diode for minimal distortion at working temperature, we take this effect ito account.

JPV

 

[Bob Cordell]

Re = Re (external) + Re (internal).
Re (internal) or parasitic emitter resistance is not know as value, is not even constant, so 26mV is not a value measured over external resistance.

You are exactly right; that is why I used the term "theoretically". The 26 mV number is the number at room temperature for an ideal output transistor.

Internal emitter resistance, or base resistance divided by beta, will reduce this number. Base resistance divided by beta can be a bit problematic because both beta and base resistance are a function of collector current.

Moreover, I suspect that many people overlook the fact that base stopper resistors will also influence the 26 mV number. For example, a 5-ohm base stopper resistor with a transistor beta of 50 will yield an equivalent resistance as seen looking into the emitter of 0.1 ohm.

Cheers,
Bob

 

[AndrewT]

Internal emitter resistance, or base resistance divided by beta, will reduce this number. Base resistance divided by beta can be a bit problematic because both beta and base resistance are a function of collector current.

Is there a way to experimentally determine (an approximate) value of internal base and emitter resistances.
Is Rbb = internal base resistance?

If using Onsemi's NJL3281 with a 2r2 base stopper, the external base resistance reflected to the emitter is only 2.2/110 ~ 0r02 for Ic between 20mA and 400mA.
Will this swamp the internal base resistance?

 

[JPV]

Is there a way to experimentally determine (an approximate) value of internal base and emitter resistances.
Is Rbb = internal base resistance?

If using Onsemi's NJL3281 with a 2r2 base stopper, the external base resistance reflected to the emitter is only 2.2/110 ~ 0r02 for Ic between 20mA and 400mA.
Will this swamp the internal base resistance?

As Bob said, Rbb is changing with the collector current. For a normal transistor, Rb increases with decreasing current which is a pitty because correct biasing ( 26mV) has to happen at low current ( at crossover).
Therefore, the only way is to monitor the distortion while adjusting bias. A good experiment would be to monitor distortion while adjusting the base stopper resistor.

JPV