It's been interesting to hear what you've discovered about Op Amps

and noise. The formula you gave for adding the different kinds of

noise is an important one.

Low-noise audio preamps are actually one of the toughest design tasks

in electronics. Op Amps are even trickier, noise-wise, because they

use differential input stages, which involve more than one

transistor. And they involve feedback networks.

You've identified the two major noise relationships: low frequency

voltage noise (caused by transistor junction self-noise), and current

noise (caused by random bias currents into the input's source

impedance). Which of the two will dominate depends largely on the

device you select, and the impedance of the driving stage or sensor.

Op Amp manufacturers usually emphasize one particular spec (voltage

noise or current noise) when marketing their devices. If you look

closely, you'll probably notice that if they are advertising a very

low voltage noise (like 1nV/sqrtHz), the current noise specs will be

HIGHER. And if they are advertising very low current noise (like

100fA/sqrtHz), the voltage noise will be higher. Unfortunately, the

things they need to do to make one lower, causes the other to be

higher. With the LT1028 you mentioned, for example, which is

marketed as a very low voltage noise device (<1nV/sqrtHz) - the

current noise is 1-1.6 pA/sqrtHz. Using a 1000 ohm input resistor,

the current noise will actually dominate (at 1KHz).

Another tricky thing about noise specs in Op Amps is where the noise

knee frequency is. Some manufacturers specify this, and some

conveniently leave it out. I recently saw a 1 GHz Op Amp (AD8003?)

specified with 1.8 nV/sqrt noise. From the marketing sheet, one

could easily be misled into thinking this would be a good device for

audio use, but the 1/f knee frequency was so high (~20 KHz) that the

audio voltage noise would be horrendous.

A lot of manufacturers specify the 1/f knee frequency as 1 KHz (for

audio-frequency low-noise op amps), but many simply don't specify it,

even when they specify the noise performance. And the knee frequency

- along with the noise spec - pretty much defines the noise

characteristic. Even for a particular device, the knee frequency can

actually vary quite a bit from device-to-device, making it an even

harder target to hit. A really good low-noise audio Op Amp (or

transistor) will have it's knee frequency specified as 100 Hz, or

lower. If you really want low noise performance, you have to also

look at the knee frequency. If looking for a low-noise audio Op Amp,

for example, first look for a knee frequency LOWER than your lowest

passband frequency. Then look for the noise spec. Unfortunately,

the knee frequencies are often buried in the fine-print

notes. However, most (but not all) manufacturers that target the

low-noise audio market also publish the low frequency voltage noise

AND current noise CURVES, so you can see for yourself what noise

levels you'll encounter at the design frequency. Keep in mind,

however, that curves are almost always TYPICAL, not spec, and based

on a particular test circuit, and actual performance can vary quite a bit.

One thing I've discovered is that second-source devices (example:

NJM's version of a National Op Amps) almost never meet the noise spec

at the same knee frequency. They're usually quite a bit cheaper, but

you end up paying for it in noise performance.

The rule-of thumb I've always heard is that for source impedances

less than 100 ohms, the voltage noise spec will be most important,

and if more than 1000 ohms, the current noise spec. And in between

those two values, it will be a balancing act. For direct conversion

detectors (like those used in the SDR), where the output impedance is

50-200 ohms, and the low-frequency (high-pass) cutoff is 100-300 Hz,

it would be best to look for devices with a 1/f knee frequency as low

as possible (definitely less than 1 KHz), and secondarily go for the

lowest current noise you can find. And always breadboard the circuit

to see if it really performs as expected.

Another aspect with Voltage Op Amps, is that you're dealing with an

input resistor and feedback resistor. If you want to minimize

current noise, you need to minimize the resistance of the input and

feedback resistors (and maybe match your input impedance), which

means the op amp has to have more current sourcing capability, which

means it probably won't approach rail-to-rail operation. Oops, there

goes your dynamic range... Another balancing act.

Good Op Amps won't be cheap, however, which is why a lot of designers

stick with discrete bipolar preamps and low impedance circuits,

sometimes paralleling devices. The trade-off there is often higher

bias currents needed for good low-noise operation, and complexity, of

course. And to add to the design headache when using the discrete

approach, the common general purpose transistors are rarely specified

for low-frequency noise performance (or have published curves for

noise in the audio range).

Just a few notes, from years of beating my head against the "noise" wall ;-)

73,

monty N5ESE

http://www.dit-dididit-dit.com

At 11:19 AM 12/31/2006, Kees & Sandy wrote:>I've been interested in all the noise concerns relative to SDR OpAmp

>use and finally found this interesting imformation in the Linear

>Technology spec sheet for the LT1028/LT1128 (page 10):

>

>The total referred noise of an OPAmp is the square root of the sum of

>the squares of voltage noise (that's in the OpAmp spec), the total

>equivalent source resistance at the two inputs, and the current noise

>(that's in the OpAmp spec) times the equivalent source resistance.

>

>Total = sqrt(en^2 + rn^2 + inReq^2)

>

>en = xx nV/sqrtHz (from the specifications)

>

>Req = source resistance + switch resistance + gain resistor

> resistance + plus any other resistance between the

> input terminals

>

>rn = 0.13 sqrt Req .....in nV/sqrtHz (for 25 degrees C)

>

>in = (yy pA/sqrtHz)Req (from specifications) [This term does not

>really come into play unless the source resistance is above several K

>ohms so, for SDR use, we can drop it.]

>

>Output noise = Gain times total input noise

>

>-------------------------------------------------------

>-------------------------------------------------------

>What this says to me is that the voltage noise spec is important, but

>the resistance noise spec is also (sometimes more) important.

>

>For example, you may have a 1nV/sqrtHz OpAmp but if you have 200 ohms

>input resistance, that contributes 1.83nV/sqrtHz, for a total of

>2.08nV/sqrtHz. If you have 100 ohms input resistance, the total would

>be 1.52nV/sqrtHz.

>

>If the total is 1.5nV/sqrtHz and the OpAmp gain is 100, that's

>150nV/sqrtHz output and if the bandwidth received is 100Khz, that

>means 47.4uV noise output if you want to calculate signal to noise

>ratios.

>

>73 Kees K5BCQ

>

>

>

>Via the Austin QRP Club list

>Yahoo! Groups Links

>

>

>