Dick Kleijer's website has an excellent article on low noise device measurements, this part of the article has a table of measured noise spectra and calculated Rbb's that's very relevant here: Ultra low noise amplifiers
AS194 AS394 ( a remake of LM194 LM394 ). Matched NPN transistor pair
Noise Voltage Density of 1,8 nV/ rt Hz
http://www.alfarzpp.lv/eng/sc/AS394CH.pdf
Noise Voltage Density of 1,8 nV/ rt Hz
http://www.alfarzpp.lv/eng/sc/AS394CH.pdf
Better quality silicon these days should make quieter parts possible than could be made back in the 80s, especially the 1/f noise
That's excellent!! It would appear, from his research, that the ZTX851 and its brothers are about the lowest noise transistors available---he's quoting voltage noise of 0.223nV√Hz for these. Anyone find anything better? How about low-noise FETs?
Last edited:
If you are willing to spend a fortune on a FET with an absurdly large die area, they can have quite low voltage noise (though not as low as 0.223 nV/sqrt(Hz)), see the InterFET IF3601 and IF3602 datasheets.
Thanks---that FET sure is quiet! Datasheet says 0.3nV√Hz. But oh MY---$21.64 EACH!!!
Is there anything else that's pretty quiet but not so much $$$$$???
Sure, a dozen AM car radio front-end JFETs in (AC) parallel. The BF862 is out of production, but there are alternatives, such as the NSVJ3557SA3 and the NSVJ5908DSG5. I haven't read it, but I know that Dimitri Danyuk wrote an article about low-noise FETs some time ago.
Last edited:
Thanks---it seems the NSVJ3557SA3 has a noise figure of 1 db @ 1 K Hz,and 1KΩ, per the datasheet. If I'm calculating correctly, that would be ~ equivalent to a 4.5nV√Hz voltage noise; AND it's only 57¢!! Alas, it's an SMD product. Is there an equivalent in a TO-92 package?
1 dB in 1 kHz corresponds to 2.036377 nV/sqrt(Hz) when there is no noise current. However, a graph in the datasheet also shows 2 dB in 100 ohm, which boils down to 0.967852 nV/sqrt(Hz). In fact the whole graph looks weird, because the noise figure approaches 1 dB with increasing source resistance and then stays there. Calibration error or incorrect noise bandwidth maybe?
I don't know any through-hole equivalents, but Linear Systems has some pretty good through-hole FETs. An On Semiconductor J109 may also be interesting for you, although its 1/f noise is a bit high.
I don't know any through-hole equivalents, but Linear Systems has some pretty good through-hole FETs. An On Semiconductor J109 may also be interesting for you, although its 1/f noise is a bit high.
Pretty sure all the good jfets are going to be smt. Onsemi's/former Sanyo are in the same ballpark as the one Marcel shared, and I believe in Dmitri's list as well.
Not sure I quite follow you here. 1/f noise is pink noise, eh? I see that the noise is quite low >500Hz, rising @ 0.75nV√Hz per octave below that.
Linear Systems offers most of their jfets in standard through hole packages.
Trendsetter sells them.
Exactly, the knee frequency is high. Ideally you'd like 1/f knee at or below 20Hz for audio, but that's pretty rare.
Noise factor/figure is ratio of amp noise to (amplified) source noise, 1dB is 1.26 ratio, so the amp acts like 260 extra ohms on top of the source 1k ohms, Johnson noise of 260 ohms is about 2nV/sqrt(Hz)
<20Hz is extremely rare, and not very important because of low bandwidth. (Also typical room rumble, and because Fletcher-Munson.)
The actual +3dB knee freq is below 100Hz which is very good for audio.
Yes, the NF should decline at high R but 1dB may just be test gear limit.
These parts are about as good as it gets for >150r source.
It is folly to be chasing dB of NF. If 3dB NF is "not good enough", after honest listening tests, you want to re-think the whole situation. Put the mike closer to the flea. Use a transformer on your MC needle. Etc.
Hans Polak and Syn08 have been messing around with the ZTX851 for MC headamps and getting some very good results. Some argument over where the 1/f knee is, but seems to be around 10Hz if you keep Ic around 5mA. In that application its low enough noise that most cartridges produce significantly more. Hfe is a little low but if you need lowest noise at a sensible price in a BJT it's the goto.
The remark about the not so low 1/f corner applied to the On Semiconductor J109, which has a knee frequency far above 100 Hz. Nonetheless it's not a big deal; even 700 Hz is fine when you look at the A- and ITU-R 468 weighting curves.
It is easy enough to subtract the noise of the test gear, for example with hot-cold methods. It amazes me that On Semiconductor didn't bother to measure the noise of the NSVJ3557SA3 more accurately, considering that its low noise appears to be its main selling point. In fact it is meant for car radio AM LNAs, and designers of those would do almost anything for a whole decibel of noise reduction.
Do you also design your main amplifiers to distort as much as the loudspeaker? In any case, the purpose of a step-up transformer for MC cartridges is to reduce the noise figure.