Hi Tad,
Thanks for reading my book.
You are right about the idea of presenting more amplifier material in the form of buildable designs, like construction projects. This is what I definitely want to do as soon as I get some time. I do have a large number of pet projects I would like to publish as well, not all of which are amplifiers. Some, for example, are like the Distortion Magnifier project published in Linear Audio. Going forward I'm going to publish more articles in various magazines, some of which will be construction projects. I also really get my enjoyment from building the designs, seeing them work, and listening to them.
I definitely do want to come out with one or more reference designs based on the design techniques covered in the book.
Thanks again for your interest.
Cheers,
Bob
Hi Olivier,
I'm not sure what Randy's thinking was, but I like the current source of 2.10f and have not had stability problems. I think it probably gives the highest output impedance for a given amount of required voltage headroom. Note that Doug Self also uses this current source in his Blameless amplifier. If one does encounter any HF stability issues, a resistor can be placed in series with the base of Q2. Bear in mind that this current source produces a current that is not temperature compensated. As temperature increases, the value of the current produced will decrease, since the Vbe of Q2 placed across R1 will decrease with increases in temperature. A current source using two silicon diodes in series to bias the base of the current source transistor will also exhibit the same type of temperature dependence.
Cheers,
Bob
Current Source
Hi Bob,
Well, actually Edmond Stuart once replaced the current source of 2.10f into one with a Zener reference. The reason for it was to keep the base voltage of the current source Q far enough away from the rail voltage which in turn leaves a higher voltage between rail and emitter. So this enables the use of a bigger resistor in series with the emitter. To stabilize for temperature ? It was like 1k ore so instead of usualy 100 or 120 ohms (0,7V for 5mA say)
Cheers,
Olivier
Hi Bob,
Well, actually Edmond Stuart once replaced the current source of 2.10f into one with a Zener reference. The reason for it was to keep the base voltage of the current source Q far enough away from the rail voltage which in turn leaves a higher voltage between rail and emitter. So this enables the use of a bigger resistor in series with the emitter. To stabilize for temperature ? It was like 1k ore so instead of usualy 100 or 120 ohms (0,7V for 5mA say)
Cheers,
Olivier
Hi Olivier,
If one is willing to spend some voltage headroom, there are numerous ways to further improve a current source. That is often the key; use of the zener diode, as in my Figure 2.10d, costs voltage headroom. One can also spend voltage headroom by cascoding a current source, like the one in either 2.10d or 2.10f.
Note that the feedback current source of 2.10f can also be improved if one is willing to spend some voltage headroom. For example, one can place a zener diode in series with the emitter of Q2, causing a zener drop plus a Vbe to be forced across the emitter resistor of Q1. This also places the emitter further from the rail and allows the use of a larger emitter resistor.
Interestingly, if you use a 5.1V zener, the positive temperature coefficient of the zener voltage will approximately cancel the negative temperature coefficient of the Vbe, resulting in a current source with relatively temperature-compensated current.
The ability to have voltage headroom to spend is a further good reason to employ boosted supplies for the IPS-VAS in a power amplifier.
Cheers,
Bob
Current Source
Hi Bob,
Interesting solutions and now I see exactly the reason for Edmonds choice. It was a 6,8V zener by the way.
Since I am going for a mirror image topology I need to know something else. If the topology isn't a mirror the 2 diodes in series as voltage reference for the current source Q are lead to ground through a resistor that sets the diode current. When this is used in a mirror image topology the reference to ground is often omitted. It is told me this is a good technique.
Is this also true for the 2.10f current source?
In my configuration your 50Vrail is near ground and your ground is at my negative rail. Upside down for the top half of the schematic. Your R2 would then connect the upper and lower Q2 together without passing through GND. Do it that way or keep them seperate referencing them to GND? What's best in this case?
I suppose temperature compensated cascoded CCS's must be overkill right? In your opinion, is temperature dependency a big issue for CCS in an amplifier? My target amp is 250W into 8ohm / L-MOSFET / THD < 0,01% @ full power & 20KHz input signal.
Off-Topic : Would you know where I could find the spice models for the KSA992 & KSC1845?
Cheers,
PS : Try Belgian Beers
Olivier
Hi Bob,
Interesting solutions and now I see exactly the reason for Edmonds choice. It was a 6,8V zener by the way.
Since I am going for a mirror image topology I need to know something else. If the topology isn't a mirror the 2 diodes in series as voltage reference for the current source Q are lead to ground through a resistor that sets the diode current. When this is used in a mirror image topology the reference to ground is often omitted. It is told me this is a good technique.
Is this also true for the 2.10f current source?
In my configuration your 50Vrail is near ground and your ground is at my negative rail. Upside down for the top half of the schematic. Your R2 would then connect the upper and lower Q2 together without passing through GND. Do it that way or keep them seperate referencing them to GND? What's best in this case?
I suppose temperature compensated cascoded CCS's must be overkill right? In your opinion, is temperature dependency a big issue for CCS in an amplifier? My target amp is 250W into 8ohm / L-MOSFET / THD < 0,01% @ full power & 20KHz input signal.
Off-Topic : Would you know where I could find the spice models for the KSA992 & KSC1845?
Cheers,
PS : Try Belgian Beers
Olivier
Hi Bob,
---Interestingly, if you use a 5.1V zener, the positive temperature coefficient of the zener voltage will approximately cancel the negative temperature coefficient of the Vbe, resulting in a current source with relatively temperature-compensated current.---
I wonder if a zener of this voltage could have an incidence on noise.
---Interestingly, if you use a 5.1V zener, the positive temperature coefficient of the zener voltage will approximately cancel the negative temperature coefficient of the Vbe, resulting in a current source with relatively temperature-compensated current.---
I wonder if a zener of this voltage could have an incidence on noise.
Hi Bob,
Interesting solutions and now I see exactly the reason for Edmonds choice. It was a 6,8V zener by the way.
Since I am going for a mirror image topology I need to know something else. If the topology isn't a mirror the 2 diodes in series as voltage reference for the current source Q are lead to ground through a resistor that sets the diode current. When this is used in a mirror image topology the reference to ground is often omitted. It is told me this is a good technique.
Is this also true for the 2.10f current source?
In my configuration your 50Vrail is near ground and your ground is at my negative rail. Upside down for the top half of the schematic. Your R2 would then connect the upper and lower Q2 together without passing through GND. Do it that way or keep them seperate referencing them to GND? What's best in this case?
I suppose temperature compensated cascoded CCS's must be overkill right? In your opinion, is temperature dependency a big issue for CCS in an amplifier? My target amp is 250W into 8ohm / L-MOSFET / THD < 0,01% @ full power & 20KHz input signal.
Off-Topic : Would you know where I could find the spice models for the KSA992 & KSC1845?
Cheers,
PS : Try Belgian Beers
Olivier
I use 2.10F exclusively. The source will hold within .05mA from outside (0c) all the way to the max hairdryer temp (60c+). With "F" gain group KSC1845's, this source has no stability issues. I also use 2sc2229's as 2.10f sources. Once , with 2sc1845 - gain group "U" , I encountered parasitic oscillation , I added a 390pF poly cap between bases. A 1k basestopper on the output device of 2.10f will do the same thing. Below are the KSC1845/sa992's models you wanted (also the ON semi models).
I have many more models in " /WEBSITE/ZIP1/Electronics%20Software/ " at my link below.
OS
Are you sure about the zener voltage? A 5.1 V typically has about a zero tempco at 5 mA. A 6.2 V is usually used for ~+2 mV/°C. Lower currents may shift these figures even more towards negative tempcos.
For some typical 6.2 V parts at 5 mA I measured about 40 nV/rtHz, so this is clearly higher than what we get from a B-E junction (say about 1 nV/rtHz). On the other hand, the larger emitter resistor is very beneficial to reduce noise. For a CCS without zener and an output current of 1 mA we need a 620 Ohm emitter resistor which contributes 3.2 nV/rtHz, so overall output current noise density is roughly sqrt(3.2 nV/rtHz^2 + 1 nV/rtHz^2)/620 Ohm = 5.4 pA/rtHz.
With a 6.2 V zener the emitter resistor becomes 6k8 which contributes 11 nV/rtHz. Output current noise density is sqrt(40 nV/rtHz^2 + 11 nV/rtHz^2 + 1 nV/rtHz^2)/6800 Ohm = 6.1 pA/rtHz. So output current noise is almost the same.
Of course these calculations aren't exact (e.g. we have ignored noise from pass transistor base current and the effect of pass transistor gm), but they are useful design guide lines.
Another consideration is how sensitive the amplifier actually is to CCS noise. For typical amplifier topologies CCS noise appears either as a common-mode signal (e.g. IPS) or is fed to a node with comparably low impedance and further reduced by global feedback (e.g. VAS). So no particularly high sensitivity is given.
Samuel
Last edited:
Hi Bob,
Interesting solutions and now I see exactly the reason for Edmonds choice. It was a 6,8V zener by the way.
Since I am going for a mirror image topology I need to know something else. If the topology isn't a mirror the 2 diodes in series as voltage reference for the current source Q are lead to ground through a resistor that sets the diode current. When this is used in a mirror image topology the reference to ground is often omitted. It is told me this is a good technique.
Is this also true for the 2.10f current source?
In my configuration your 50Vrail is near ground and your ground is at my negative rail. Upside down for the top half of the schematic. Your R2 would then connect the upper and lower Q2 together without passing through GND. Do it that way or keep them seperate referencing them to GND? What's best in this case?
I suppose temperature compensated cascoded CCS's must be overkill right? In your opinion, is temperature dependency a big issue for CCS in an amplifier? My target amp is 250W into 8ohm / L-MOSFET / THD < 0,01% @ full power & 20KHz input signal.
Off-Topic : Would you know where I could find the spice models for the KSA992 & KSC1845?
Cheers,
PS : Try Belgian Beers
Olivier
Hi Olivier,
I'm guessing it makes little difference whether the current source biasing resistors in a complementary IPS/VAS connect to ground in the middle or not. However, by what you have said I am assuming you are talking about the current sources for the input LTPs, since most complementary VASs don't employ a current source. Anyway, the current sources of 2.10f are relatively insensitive to the biasing current and thus the rail voltage. I suppose that if the rail voltages were different, one could argue there would be a slight difference in the tail currents of the complementary LTPs, but I think other factors might contribute more to differences, such as differences between the NPN and PNP Q2 device's Vbe in the two polarities of current sources. Systematic differences in these could be compensated by a very slight difference in choice of the emitter resistors if one really cared that much about it.
I hope I'm not mis-understanding your question about the current sources in a complementary design, but bear in mind that if you are just talking about current sources for the input LTPs, you likely will not have a voltage headroom concern and can afford to waste voltage headroom in pursuit of better current source performance. In that case, you actually might want to go with a cascoded current source, and maybe one that uses a zener as well.
Cheers,
Bob
Current Source
Hi Bob,
I am indeed talking about the IPS current sources. I will probably go for the 2.10f solution w/o reference to ground and also use a 1K basestopper resistor.
Zenering and cascoding them might be an option if the project works with the 'simple' 2.10f solution.
I am also wondering if it's an option for me to cascode the IPS itself or not? But I think I will start off w/o cascode. Same as for the current sources, they can be added later if the basic works.
Cheers,
Olivier
Hi Bob,
I am indeed talking about the IPS current sources. I will probably go for the 2.10f solution w/o reference to ground and also use a 1K basestopper resistor.
Zenering and cascoding them might be an option if the project works with the 'simple' 2.10f solution.
I am also wondering if it's an option for me to cascode the IPS itself or not? But I think I will start off w/o cascode. Same as for the current sources, they can be added later if the basic works.
Cheers,
Olivier
Capacitors
Hi Bob,
I noticed in your book that you don't use/show capacitors e.g across zener references or any other similar use. Also the current source of 2.10f is not shown with a capacitor to somewhat filter rail variations.
Are they just left out for simplified lecture or do you think this is up to the reader to decide. If the latter is the case I presume there might be reason not to use capacitors there. Is that so? Or is it rather 'more than desireable but not really a must'?
Referencing the 2.10f to Randy's drawings he uses to split the Q2 resistor in half and places a capacitor from between the resistors to the power rail.
Would there be a reason not to place them? I will not use top notch capacitors there so maybe none is better than plain standard electrolytic capacitors?
Cheers,
Olivier
Hi Bob,
I noticed in your book that you don't use/show capacitors e.g across zener references or any other similar use. Also the current source of 2.10f is not shown with a capacitor to somewhat filter rail variations.
Are they just left out for simplified lecture or do you think this is up to the reader to decide. If the latter is the case I presume there might be reason not to use capacitors there. Is that so? Or is it rather 'more than desireable but not really a must'?
Referencing the 2.10f to Randy's drawings he uses to split the Q2 resistor in half and places a capacitor from between the resistors to the power rail.
Would there be a reason not to place them? I will not use top notch capacitors there so maybe none is better than plain standard electrolytic capacitors?
Cheers,
Olivier
Hi Samuel,
---For some typical 6.2 V parts at 5 mA I measured about 40 nV/rtHz, so this is clearly higher than what we get from a B-E junction (say about 1 nV/rtHz). On the other hand, the larger emitter resistor is very beneficial to reduce noise. For a CCS without zener and an output current of 1 mA we need a 620 Ohm emitter resistor which contributes 3.2 nV/rtHz, so overall output current noise density is roughly sqrt(3.2 nV/rtHz^2 + 1 nV/rtHz^2)/620 Ohm = 5.4 pA/rtHz.
With a 6.2 V zener the emitter resistor becomes 6k8 which contributes 11 nV/rtHz. Output current noise density is sqrt(40 nV/rtHz^2 + 11 nV/rtHz^2 + 1 nV/rtHz^2)/6800 Ohm = 6.1 pA/rtHz. So output current noise is almost the same.
Of course these calculations aren't exact (e.g. we have ignored noise from pass transistor base current and the effect of pass transistor gm), but they are useful design guide lines.
Another consideration is how sensitive the amplifier actually is to CCS noise. For typical amplifier topologies CCS noise appears either as a common-mode signal (e.g. IPS) or is fed to a node with comparably low impedance and further reduced by global feedback (e.g. VAS). So no particularly high sensitivity is given.---
Thanks for your comments.
So, concerning noise, zener diodes have much worse press than they merit. Having a simple feedback CCS with reduced tempco can sometimes be desirable.
I think a feedback CCS using a TL431 would have a stable output current with temperature but is probably more noisy than the above ones and its impedance output may not be as good at very high frequencies.
---For some typical 6.2 V parts at 5 mA I measured about 40 nV/rtHz, so this is clearly higher than what we get from a B-E junction (say about 1 nV/rtHz). On the other hand, the larger emitter resistor is very beneficial to reduce noise. For a CCS without zener and an output current of 1 mA we need a 620 Ohm emitter resistor which contributes 3.2 nV/rtHz, so overall output current noise density is roughly sqrt(3.2 nV/rtHz^2 + 1 nV/rtHz^2)/620 Ohm = 5.4 pA/rtHz.
With a 6.2 V zener the emitter resistor becomes 6k8 which contributes 11 nV/rtHz. Output current noise density is sqrt(40 nV/rtHz^2 + 11 nV/rtHz^2 + 1 nV/rtHz^2)/6800 Ohm = 6.1 pA/rtHz. So output current noise is almost the same.
Of course these calculations aren't exact (e.g. we have ignored noise from pass transistor base current and the effect of pass transistor gm), but they are useful design guide lines.
Another consideration is how sensitive the amplifier actually is to CCS noise. For typical amplifier topologies CCS noise appears either as a common-mode signal (e.g. IPS) or is fed to a node with comparably low impedance and further reduced by global feedback (e.g. VAS). So no particularly high sensitivity is given.---
Thanks for your comments.
So, concerning noise, zener diodes have much worse press than they merit. Having a simple feedback CCS with reduced tempco can sometimes be desirable.
I think a feedback CCS using a TL431 would have a stable output current with temperature but is probably more noisy than the above ones and its impedance output may not be as good at very high frequencies.
Yes and no. At lower currents than 5 mA and breakdown voltages above 6.2 V noise quickly increases drastically. If we replace the 6.2 V zener in the above example by say 5 forward biased diodes output noise will drop a lot--at the cost of temperature stability and supply rejection.
The important thing to know about current source noise is that the emitter/source resistor needs to be large for a given output current, i.e. the DC voltage across it should be large. Low noise and low saturation voltage are difficult and at some point even impossible to combine.
For a self-biased JFET current source we want a JFET with high Vp as this gives a large resistor value. 2KS170 et al are a poor choice; get a J111. Even if the basic voltage noise performance of the J111 is say 3x worse the typically more than 10x larger source resistor easily makes up for this.
Samuel
Last edited:
Hi forr,
I always assume zeners will have some noise, so I always bypass them with an electrolytic.
Cheers,
Bob
Hi Samuel,
These are really good points. Better immunity to CCS noise is another good reason for using an input circuit design that has good common mode rejection, as you point out.
The zener temperature coefficient I used was from the attached Diodes Inc Zener data sheet, where tempco is shown as about +0.038%/C for a 5.6V zener at the test current of 20mA. I honestly have not studied how much this positive tempco would decrease at lower operating current, so you may have a point. Do you have data that shows the zener tempco to be so strong a function of operating current as to make it go close to zero at operating currents in the 2-5 mA range?
Cheers,
Bob
bypassing at the Zener is not the best idea - the Zener is supposed to have a low dynamic resistance when biased enough to regulate - so the R in the RC filter constant is low - better to add a discrete RC filter
if trying to stabilize the Zener V with respect to PS V ripple the bias R should be split and the midpoint bypassed, if trying to lower ouput noise then a RC should be between the Zener and the load (ccs Q base)
diff pair tail current noise is not too important with mirror load - noise 1st order differences out
likewise stability isn't too important as long as variation is small and slower than audio frequencies - tempco isn't often a 1st order audio ccs concern
if trying to stabilize the Zener V with respect to PS V ripple the bias R should be split and the midpoint bypassed, if trying to lower ouput noise then a RC should be between the Zener and the load (ccs Q base)
diff pair tail current noise is not too important with mirror load - noise 1st order differences out
likewise stability isn't too important as long as variation is small and slower than audio frequencies - tempco isn't often a 1st order audio ccs concern
www.nxp.com/documents/data_sheet/BZV55_SER.pdf see page 7, bottom.
I believe that the wild variation of tempco with bias condition is a result of the mixed breakdown mechanism of the lower voltage parts. Zener effect has a negative tempco, avalanche a positive (see Zener diode - Wikipedia, the free encyclopedia). Higher bias currents seem to shift the diode towards avalanche breakdown.
Consider the required value.
Samuel