Quieter voltage reference: LED or LM329AH?

I have built current sources using LEDs to bias the base of a BJT in the past. This is what I thought I would use in my current project, a BoSoZ preamp. But in my last order from Digikey I splurged on some LM329AH buried zener voltage references.

The LM329 is a handy little device with terrific specs, but there is one thing I am unable to determine. The LM329 is specified for 7µv RMS noise, but I haven't been able to find any specification for the noise of an LED. Assuming that the current through the LED is tightly regulated, how would the noise from an LED reference compare to a buried zener reference?

Any noise induced by the voltage reference in the current source will be translated directly to the output, so this is an important factor for making the kind of dead-silent idle signal I desire. I'm assuming the buried zener is quieter but I'm not equipped to measure that here, and my breadboard version of the preamp is only allowed to drive a really crappy bench speaker, so I can't tell by listening either.

By the way, the LEDs I use are Panasonic LN424RP that produce a 1.8V drop when you pull 1mA through them with a CCS e.g. Vishay J505, and are also rather dim, which is a bonus in my book.
Thanks for the pointer to better parts, but what about the question at hand? How does the noise compare to noise from LED reference?

Frankly I could not care less about initial accuracy, temperature drift, and so forth, because these effects are not fast enough to induce audible noise into the signal.
LED noise


According to Sonny:

diyAudio Elder

Registered: Oct 2001
Location: Denmark

I don't want to fight with you!... LED's do generate noise .. take it leave it . It is still a PN design. It is right it does not generate a lot of voltagenoise.

Its series resistance applies to the diodeformula : ~ re = 25mV/A (at room temperature).

At a current of 100uA it will generate : ~2nV/sqroot(Hz)
At a current of 1mA it will generate : ~0.6nV/sqroot(Hz)
At a current of 10mA it will generate : ~0.2nV/sqroot(Hz)

Yes it is VERY small values .. Every time you raise the current your currentnoise will go up and it will multiply it self with the series resistor .. Wich also generate thermal related noise ...
I dont know the value of the currentnoise.. I have not been able to find any noisefigures on LED's at all!! It is not the normal application for LED's

The biggest reason i can think of why use a LED as refence is that it have the same Tempco as a transistor and to keep the circuit simple.

JFET do have noise figure around 3nV (For the J113). If you look at the currentnoise it is a lot lower than a BJT .. maybe 100times!
So you only end up with the 3nV and the thermal noise generated by the resistors and caps! surrounding the JFet.

If you choose the right JFET (in my case J113) to a specifiet current you end up with a ZERO tempco. Even if it would change 3% i would be happy .. The LED - transistor circuit can not do this much better... They say 2% over a complete temp. range"

I use green LEDs (1.8V) and a J510 for bias. The LM329 can biased down to a little below a milliamp. Both are excellent low noise references for analog circuits. The tempco for the LEDs are not as good but I don't need that precise of a supply voltage for audio circuits. I hope to measure the PSRR of the LED J501 this week and report back on the results.


Monte McGuire

Paid Member
2013-07-28 2:58 am
The only gotcha to be aware of is that LEDs make good photosensors. If you allow fluorescent or HF driven LED light to shine on them, you'll get switching noise injected into your circuit. Dip them in opaque paint before you solder them in, and you'll avoid this problem.