Lowest noise BJT transistor?

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Thanks again for the answer.
Provided I am using the 2SA1085 and complementary with no feedback, could you show me numerically how you get to the ideal conditions?
I am so sorry to bother you with these questions, but I find the subject very interesting, and I would like to make sure I am choosing the right operating conditions.
I am cascading the BJTs setting VCE=12V. I have 5R2 emitter resistors. two BJTs in parallel for each halve of the differential input.

When you talk about internal base resistance, are you referring to the rbb parameter?
If so, where you do you determine it? If there is simulation model from manufacturer, then it is in there (assuming they have it created it properly) otherwise how else do you determine that?
 
Thanks again for the answer.
Provided I am using the 2SA1085 and complementary with no feedback, could you show me numerically how you get to the ideal conditions?
I am so sorry to bother you with these questions, but I find the subject very interesting, and I would like to make sure I am choosing the right operating conditions.
I am cascading the BJTs setting VCE=12V. I have 5R2 emitter resistors. two BJTs in parallel for each halve of the differential input.

When you talk about internal base resistance, are you referring to the rbb parameter?
If so, where you do you determine it? If there is simulation model from manufacturer, then it is in there (assuming they have it created it properly) otherwise how else do you determine that?

If specified, you can estimate rbb from the equivalent input noise voltage at a relatively high collector current and zero source impedance. The datasheet of the 2SA1085 states that the equivalent input voltage noise density is 0.5 nV/sqrt(Hz) at 10 mA of collector current. When you square this and divide it by 4kT, you find that this is equivalent to the thermal noise voltage of a resistor of about 15 ohm.

You then have to correct for the collector shot noise. It can be shown that you have to subtract kT/(2*q*Ic), which is about 1.3 ohm when Ic is 10 mA. Hence, rbb is roughly 13.7 ohm.

So, given a 33.3333... ohm source impedance, 13.7 ohm base resistance and 5.2 ohm in the emitter lead, at hFE ~=400, the optimal collector current for a single common-emitter stage would be sqrt(hFE) * 26 mV/(33.3333... ohm + 13.7 ohm + 5.2 ohm) ~= 10 mA.

However, you are using a differential pair and you are connecting two devices in parallel. Using a differential pair increases the equivalent input noise voltage by sqrt(2) and reduces the equivalent input noise current by sqrt(2) (assuming that your circuit is fully balanced, otherwise it is more complicated). Using two devices in parallel does exactly the opposite. All in all, the collector current that is optimal for a single common-emitter stage should also be optimal for each transistor in your configuration.

So, the optimum is roughly 10 mA of collector current per transistor.

The maximum power dissipation of the 2SA1085 is 400 mW at 25 degrees C and the maximum junction temperature is 150 degrees C, which means that 400 mW causes 125 degrees of self-heating. At VCE = 12 V and IC = 10 mA, you then get 37.5 degrees of self-heating. On the kelvin scale, this is still small compared to the ambient temperature.

By the way, noise optima are usually rather broad, so don't worry if the current isn't accurate. The difference in noise between 5, 10 and 20 mA is probably barely measurable.
 
  • Like
Reactions: 1 user
Marcel...I have a question for you!
How do you compare these two devices: 2SA1085 and 2SA1312 (and relative complementary) ?

They both seem to be extremely low noise, but the good thing is that the latter is still in production.
Also, please correct me if I am wrong, but from the characteristics I can see on the datasheet, the 1312 seems to be a bit more linear than the 1085, but it's hard to say that for sure given scales are different.

Could you also be so kind to comment on the noise chart?
I am still struggling very much to fully understand the noise figure.

Thanks a lot!
 
By definition, the noise figure is the difference in dB between the noise of the source and the amplifier or amplifying device together and the source on its own, assuming that the source has a noise temperature of 290 K.

Regarding the 2SA1312 datasheet, the noise figure in the table doesn't tell you much, because it is specified at a low collector current and high source resistance.

The graph shows you that at 1 kHz, the transistor can reach a 3 dB noise figure at source impedances down to 25 ohm when optimally biased. 3 dB noise figure means that the transistor generates the same amount of noise as the source. Assuming that 1/f noise is negligible at 1 kHz and correcting for collector and base shot noise, the base resistance is 25 ohm - kT/(qIc) ~= 22.4 ohm. That's a pretty good figure, even though it is not quite as good as the 2SA1085. The hFE (ratio between the power spectral densities of the collector and base shot noise) is similar to the 2SA1085.

Note that the maximum power dissipation of the 2SA1312 is much smaller than the maximum power dissipation of the 2SA1085, so it would help if you could reduce the VCE to keep the self-heating (which increases thermal noise) under control. Can you add a cascode or so? The noise optima that I've calculated are all neglecting the effect of self-heating.

If you solve the self-heating issue and use three devices in parallel, each biased at 7 mA, you get nearly the same performance as with two 2SA1085s that are each biased at 10 mA.
 
  • Like
Reactions: 1 user
In my current situation I have 2 2SA1085 in parallel per side (balanced mode) each biased at about 5.5mA with 12V across CE which is 66mW dissipation.

Unfortunately I can't increase this input stage bias current as it would dramatically affect all the power dissipation on the other stages around it, which has been really critical to nail down to obtain a stable DC performance and fit everything in the constraint space and having proper route at the same time to ensure minimum signal path and noise.

Basically if I replace part per part, while keeping same bias condition and VCE across it, will you think the replacement part will offer comparable noise performance?
 
Like I wrote earlier, noise optima are rather flat so the difference between 5.5 and 10 mA will be hard to measure. You just have to make sure you are in the right ballpark.

Besides, my calculation is based on the assumption that the junction temperature is somehow kept constant. In reality it will increase with increasing current, so the thermal noise of the base resistance is also likely to increase with increasing current (also because of the temperature coefficient of the base resistance). This will shift the optimum to a lower collector current.

You can see quite well how flat noise optima are by looking at the graph with noise figures at 1 kHz in the 2SA1312 datasheet. At 40 ohm source resistance, VEC = 6 V and Ta = 25 degrees C, the noise figure is between 1.7 dB and 2 dB for any collector current between 2 mA and 10 mA (and probably well above 10 mA, but the graph stops at 10 mA).

The 2SA1312 has a somewhat higher base resistance and a higher thermal resistance than the 2SA1085, so your noise performance will get somewhat worse with 2SA1312 transistors than with 2SA1085 transistors. It will still be quite good, though.
 
It sounds perfectly logical, but still, at 1 kHz, a 2SC2547 can reach 0.5 dB noise figure down to 70 ohm source resistance and 2 dB down to 20 ohm, while its PNP complement 2SA1085 can reach 0.5 dB noise figure down to 110 ohm source resistance and 2 dB down to 22 ohm (all at 6 V, 1 kHz and according to the datasheet plots).

On the other hand, for BC550/BC560 the trend is indeed as you would expect: the PNP has substantially less base resistance than the NPN. From the top of my head 230 ohm for the NPN and 140 ohm for the PNP, calculated from noise figure plots in old Philips datasheets.
 
Noise but the question is...

To clarify things a bit you have to take care about what type of noise you want to mitigate. Let me suggest that for transistors you care about the current noise most. Semiconductor companies often give a not-so-clear noise specification. The method you see most is a measurement of dB noise added to the signal, while using a limited bandwidth. For low frequency stuff as we do here that's the method to go for. To compare apples with apples you have to take care that noise grows with bandwidth increase, in a relation to the square root of the bandwidth growth. So a noise contribution from 100Hz to 200Hz is lower than from 100Hz to 400Hz. Wideband = more noisey. Different semi companies do specification in an extremely different way... nasty problem...:confused:
Another thing not to mix up is the noise performance for LF devices v.s. RF devices. Modern RF transistor go as low in noise to almost 0,5dB. But hey, that's for 1800MHz only. For LF this will be a useless transistor. It probably will oscillate at some frequency you don't know, and burn away from your circuit. Watch out with these devices until you know how to tame them.

I was surfing around in my endless search for these low noise transistors. Until today, and possible to purchase, I came to the following:

FET: NxP BF862 JFET
NPN: 2N5089 or MMBT5089 in SMD, 2dB noise in 10Hz to 15kHz with a source resistor of 10k.
PNP: Don't know (yet) because all input stages mostly are NPN anyway ;-)

Now we are talking transistors with decent specs and everybody can derive what the circuit performance will be. Best is to use Spice with a noise analysis, of course.

Hope this reply will help a couple of people here. :)
 
5089 has a high knee frequency where the noise increases substantially below that frequency.
Typically the plots show the knee @ 500Hz to 10kHz.
That makes it pretty noisy in Audio frequency circuits.

This indeed is something to consider: I am certain that the 1/f knee is what you mean here. But with proper choice of transistor current Ic and source resistance Rs one certainly wille be able to use this transistor as a very good imput amplifier. BTW for audio an 1/f knee of about 100Hz is not that dramatic. Unfortunately the original specs do not show us the 1/f curve (as like a MAT02 spec). But hey, compare 4 cents for a 2N5089 to the MAT02, that's a hell of a difference :)
For completeness I added the noise plot that ON-semi provided in their datasheet:
 

Attachments

  • 2N5089_NoisePlot.png
    2N5089_NoisePlot.png
    129.3 KB · Views: 1,157
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.