Hello, I note that many of Toshiba's low noise transistor pairs have their emitters joined together in the package. This seems to rule out the possibility of linearisation using emitter degeneration resistors. Does anybody know of any alternative linearisation techniques that could be used? Or alternatively, does anybody have some design tips in using these transistor pairs on input stages to reduce distortion levels?
There's the Voorman patent, which has long passed its due by date: connect the emitters of four more transistors to the same node, connect their bases to the average of the base voltages and throw away their collector current. All six transistors do need to match and to have about the same temperature for this to work. See figure 4B of EP0234655B1, https://patentimages.storage.googleapis.com/ec/5a/83/2da680180b56de/EP0234655B1.pdf
Last edited:
Another one is the multitanh: use a pair with 1:n and a pair with n:1 scaling and connect their inputs and outputs in parallel. As those dual transistors you wrote about undoubtedly have 1:1 scaling, this again requires extra transistors that still need to match the ones in the dual transistor.
Third option: use some sort of local shunt feedback. It doesn't behave similar to a degenerated differential pair anymore, then.
Third option: use some sort of local shunt feedback. It doesn't behave similar to a degenerated differential pair anymore, then.
Your 3rd option seems the most practical. Can you please elaborate? How could that be done without upsetting the balance?
On second thoughts, I don't think my third option will do much. As you want low voltage noise, both sides of the differential pair are bound to be driven from low impedances, which means that a local shunt feedback with impedances between collectors and bases will have almost no loop gain.
If Toshiba creted them , theres most probably no need for linearization.
Other than that maybe using different collector resistors whose values arent really hard to find is the simplest solution and there are a few preamps that used that method for input j-fets...just use a potentiometer with double the value of each collector transistors, find the values, then put fixed resistors in place.
Check Kenwood L-A1 for differential bootstrap transistor base current linearization technique if you wand to go even deeper cause you might probably consider a cascode too.
Having a differential cascode makes linearization less needed cause each cascode gain is much higher than just ones transistor , speeds are much higher and overall feedback way more effective.Acually in a differential pair you want the feedback transistor to have more transconductance and theres a very famous phono preamp( https://www.kenrockwell.com/audio/apt/images/holman-preamplifier/D3S_5295-bottom-2400.jpg) using an input fet and a feedbck bjt in the differential pair.Thus you can use the transistor with the higher gain(usually lower voltage noise too) in the input and the other in the feedback path.
Other than that maybe using different collector resistors whose values arent really hard to find is the simplest solution and there are a few preamps that used that method for input j-fets...just use a potentiometer with double the value of each collector transistors, find the values, then put fixed resistors in place.
Check Kenwood L-A1 for differential bootstrap transistor base current linearization technique if you wand to go even deeper cause you might probably consider a cascode too.
Having a differential cascode makes linearization less needed cause each cascode gain is much higher than just ones transistor , speeds are much higher and overall feedback way more effective.Acually in a differential pair you want the feedback transistor to have more transconductance and theres a very famous phono preamp( https://www.kenrockwell.com/audio/apt/images/holman-preamplifier/D3S_5295-bottom-2400.jpg) using an input fet and a feedbck bjt in the differential pair.Thus you can use the transistor with the higher gain(usually lower voltage noise too) in the input and the other in the feedback path.
Last edited:
Connect each BJT pair in parallel, use it as a double die-area device (higher capacitance, lower noise), and add whatever emitter resistance makes the most sense in your application. If there is a need to keep capacitance low, add cascoding.
Member
Joined 2009
Paid Member
It’s very simple, all you need to do is run them at lower current, e.g. <0.4uA each junction, which is equivalent transconductance to a higher current with added degeneration resistors. You will want to use a VAS design that avoids presenting the LTP with any large current demands, which means no large Cdom caps (add lead comp.). Using a decent feedback factor which minimizes signal swing at the differential pair Can be a good idea anyhow. Voila.
Last edited:
Two more alternatives, at least one of which is probably useless:
1. Voltage mirror
When you connect each collector to a stack of n diode-connected (base shorted to collector) transistors to the supply, you get a gain of n and the non-linearity of the current-to-voltage transfer of the diode-connected transistors compensates for the non-linearity of the voltage-to-current transfer of the differential pair. When n is large enough, the base resistance and matching requirements of the diode-connected transistors is less than the requirements on the differential pair, so maybe you can get away with BC847M or similar matched transistors that don't have connected emitters for the diodes.
However, maybe you might then just as well use the n BC847M's for the input differential pair, that will reduce the base resistance by a factor of n and reduce the RMS value of the base resistance's noise voltage by sqrt(n).
2. Multitanh with unequal junction temperatures
This will only work if the dual transistors have separate dies that are not thermally coupled very well and if the tail current is pretty high.
Use two differential pairs with 1:1 scaling. In one pair, bias the left transistor at a much higher VCE than the right transistor and in the other pair, bias the right transistor at a much higher VCE than the left transistor. You can do this by using a cascode on one output of each pair and not on the other.
Try to choose the difference in dissipation such that one die of each differential pair becomes about 24 K hotter than the other die. This results in approximately a 4:1 and 1:4 ratio of the Is values, so you can make a multitanh stage without needing to scale the transistor sizes. Experimentally optimize the temperature difference for minimal distortion.
1. Voltage mirror
When you connect each collector to a stack of n diode-connected (base shorted to collector) transistors to the supply, you get a gain of n and the non-linearity of the current-to-voltage transfer of the diode-connected transistors compensates for the non-linearity of the voltage-to-current transfer of the differential pair. When n is large enough, the base resistance and matching requirements of the diode-connected transistors is less than the requirements on the differential pair, so maybe you can get away with BC847M or similar matched transistors that don't have connected emitters for the diodes.
However, maybe you might then just as well use the n BC847M's for the input differential pair, that will reduce the base resistance by a factor of n and reduce the RMS value of the base resistance's noise voltage by sqrt(n).
2. Multitanh with unequal junction temperatures
This will only work if the dual transistors have separate dies that are not thermally coupled very well and if the tail current is pretty high.
Use two differential pairs with 1:1 scaling. In one pair, bias the left transistor at a much higher VCE than the right transistor and in the other pair, bias the right transistor at a much higher VCE than the left transistor. You can do this by using a cascode on one output of each pair and not on the other.
Try to choose the difference in dissipation such that one die of each differential pair becomes about 24 K hotter than the other die. This results in approximately a 4:1 and 1:4 ratio of the Is values, so you can make a multitanh stage without needing to scale the transistor sizes. Experimentally optimize the temperature difference for minimal distortion.
Why would that help? You still get a tanh(vin/(2 kT/q))-shaped voltage to current transfer.
Interesting, I re-invented that when I designed my first moving-magnet phono amplifier. I used a J310 for the input and a 2SC2547 for the feedback, because it led to a better noise match than using the same on both sides. Linearity would have been better with two equal JFETs, but I cared more about noise.
Member
Joined 2009
Paid Member
At it’s heart the LTP is a tanh characteristic amplifier, nothing much you can do about that. Degeneration is a local form of feedback but you can use loop feedback instead, as I suggested, to achieve similar overall results in terms of amplifier linearity.
Restricting use to this integrated LTP device will require a compromise and I prefer simplicity so my chosen compromise is low current which exacerbates the internal emitter resistance.
Restricting use to this integrated LTP device will require a compromise and I prefer simplicity so my chosen compromise is low current which exacerbates the internal emitter resistance.
Last edited:
How does that work? I've seen a few designs where one of the collector resistors is omitted entirely but I'm not sure why.
According to Mr Leach's bound [1] increasing tail current would be what you'd want to do for the lowest Re values. So running them at stupidly high collector currents might be a solution, but then you'd probably loose the low noise aspect.
[1] Amplifier Input Stage Design Criterion for the Suppression of Dynamic Distortions. W. Marshall Leach, Jr. d. Audio Eng. Soc., Vol. 29, No. 4, 1981 April.
I think jcarr's point is that as long as you don't have too stringent offset requirements, you can simply use a dual transistor as a single transistor. Using two dual transistors, each connected as a single transistor, you can again make whatever configuration you wanted to make.
You can either not use the second part of the dual transistor at all or connect it in parallel with the first part. Connecting both devices in a dual transistor in parallel halves the base spreading resistance and reduces the base spreading resistance's noise.
Years ago, I used the LM377 (~ LM1877) for things other than an amplifier, but that chip is not stable anywhere near unity gain, so I just put a small resistor (~100 Ohms) across the two inputs. This does not create a low-Z input because the resistor is ~bootstrapped. Later, I used the same idea to stabilize a similar bridged chip amp in a small stereo. Ironically, too much feedback creates distortion as a result of instabilities.
Last edited:
Did you post that in the wrong thread, steveu? (A disadvantage is the large equivalent input noise current that you get with that extra resistor, by the way.)
