LM358/LM324 DC noise (<10Hz)

[npelov]

Hi,

I'm trying to find out the limits of LM358 when it comes to noise. Most datasheets don't show noise figures. Some do, but only noise density for higher frequencies (>1kHz) and I'm interested more in DC or slow changing signals (like <10Hz). According to the TI LM385 datasheet the noise density is 40nV/Hz which is about the same as microchip's MCP606 - 38nV/Hz. What I'm really interested in is the p-p DC noise which is 2uV for MCP606. What circuit can I use to measure it?

[danadak]

These might help -

http://www.gellerlabs.com/nbw%20geller%20tech%20note.htm

Also attached.


Regards, Dana.

[T3sl4co1l]

AoE3 pg. 531 shows integrated noise, 1mHz-10Hz, LM358, about 400nV RMS.  Rising above ~100Hz (i.e., suggesting that the 1/f corner is around 100Hz, and the noise density (above there) is ca. 50nV/rtHz.

Table 8.3a (pg. 522) gives 1.8uV (0.1-10Hz, noise pk-pk) and i_n = 0.12 pA/rtHz.

Fig. 8.58 (pg. 526) shows the best noise factor is for a source impedance around 100k-1M ohm.  Which, given the DC offset on such a resistance, due to DC input bias, gives you an idea how awful this thing is...

Tim

[npelov]

1.8uV is even lower than MCP 2.8 uV p-p (typical). is this noise multiplied by gain. Also I compared TL071 which claims low noise, to TI LM358. 18 for TL071 to 40 for LM358 is not much of  a difference. Let's say you want to amplify audio signal of 16kHz. That's 18*4 = 72 for TL071 to 40*4=160 nV which is only about twice as big.

About the input current noise - it's effect on the output depends on resistance connected to inputs - gain resistors, signal output impedance, etc, right?

[Zero999]

Hi,

I'm trying to find out the limits of LM358 when it comes to noise. Most datasheets don't show noise figures. Some do, but only noise density for higher frequencies (>1kHz) and I'm interested more in DC or slow changing signals (like <10Hz). According to the TI LM385 datasheet the noise density is 40nV/Hz which is about the same as microchip's MCP606 - 38nV/Hz. What I'm really interested in is the p-p DC noise which is 2uV for MCP606. What circuit can I use to measure it?

You can reduce the noise by connecting multiple op-amps in parallel.
http://www.linear.com/solutions/5657

file:///C:/Users/Dana/Documents/Analog/Noise/Use-resistor-noise-to-characterize-a-low-noise-amplifier.pdf

That doesn't work. You need to attach files to post material on your hard drive.

[T3sl4co1l]

1.8uV is even lower than MCP 2.8 uV p-p (typical). is this noise multiplied by gain.

Yes, or more precisely, multiplied by the noise gain (which is simply the feedback ratio).  There's an Analog appnote on it, you should search for.

Also I compared TL071 which claims low noise, to TI LM358. 18 for TL071 to 40 for LM358 is not much of  a difference. Let's say you want to amplify audio signal of 16kHz. That's 18*4 = 72 for TL071 to 40*4=160 nV which is only about twice as big.

1. You should include units to make these calculations harder to get wrong;
2. You didn't take the sqrt correctly. 
(3. "Amplify an audio signal of 16kHz" doesn't necessarily mean the total bandwidth is 16kHz exactly.  It can be less.  Or will probably be more in a practical case!)

Writing it out,
18nV/rtHz * sqrt(16000Hz) = 18nV/rtHz * 126.49 rtHz = 2276 nV * rtHz/rtHz = 2.3 uV.
Rather than taking the sqrt again, the noise for the other amp is simply the ratio,
2.3 uV * (40 / 18) ( (nV/rtHz) / (nV/rtHz) ) = 5.1 uV.

Still not much, for many purposes (it's plenty stable for a control loop, say, and comparable to the noise from a similarly cheap voltage reference), but you could go for a gain as high as 100 in such a stage (because the fT is ~low MHz), bringing the output noise to nearly a millivolt (for the LM358 case).  Needless to say, this would make a poor phonograph preamplifier!

The JFET amp isn't bad, but its relatively high noise, and extremely low current noise, strongly suggests it for high impedance applications (megohms!).  In that case, its noise factor can be quite good indeed.

But even without a high impedance source, it definitely outperforms LM358.

There are better of both kinds, of course.  The takeaway is to match the amp's input impedance (in terms of noise impedance e_n / i_n) to the source, at least broadly (within a factor of 2 for NF < 3dB, I think?).

About the input current noise - it's effect on the output depends on resistance connected to inputs - gain resistors, signal output impedance, etc, right?

Yes.  And for DC bias reasons, you want the Thevenin equivalent DC resistance, at each input pin, to be equal.  (For some amps, you want the AC resistance to be matched as well: LT1028's input current noise is somewhat correlated between the two pins, so you cancel out the correlated noise by using equal AC resistances.)

Tim

[Kleinstein]

The LM358 is BJT based and has not that much 1/f noise. Though not a good BJT OP, the low frequency (like 10 mHz - 1 Hz) noise is not that bad for the LM358. There are other BJT based OPs with lower voltage noise, but often with more current noise than. So don't compare the LM358 to an LT1028 ( high current noise) but with those of comparable current noise (e.g. OP07, LT1013 might be in a similar range). With s source impedance that fits the LM358 (e.g. 100s of kOhms) the combined noise of OPs made for lower impedance sources (like the LT1028) can be higher than that of the LM358. With BJT based OPs one has look at current noise too.

Mosfest based OPs like the MCP606 on contrast usually have a rather large 1/f noise and thus give usually poor LF noise figures. So CMOS OPs generally have a higher ration of LF noise to 1 kHz noise compared to BT based OPs. On the up side there is very little current noise for FET based OPs - so they are good for high impedance sources.

[Audioguru]

The LM324 and LM358 are cheap and were not made for audio. In addition to noise they also produce awful crossover distortion and have trouble producing high levels above only 2kHz.
If you want low noise then use a low noise audio opamp.

[David Hess]

The LT1006/LT1013/LT1014 are relatively low cost lower noise improved replacements for the LM358/LM324.  There are other more modern operational amplifiers which are better yet.  Another alternative for low noise design would be to add a discrete differential preamplifier to the LM358.

If I were going to measure the noise of an LM358, I would first configure it to amplify its own noise by like 100 or 1000, use AC coupling where necessary to keep the output offset voltage in range, and then measure the output in one of a couple of ways.

If the output noise is sufficiently high after amplification, then a DSO with FFT ability should be able to make a direct measurement of noise density all the way down to almost DC depending on the FFT length.  Beware that using and interpreting the DSO's FFT results will require care.

With my 7A22 and its selectable low and high pass cutoff filters, I could use an analog oscilloscope to make an RMS spot noise measurement and then calculate the high frequency noise density.

If I only care about noise from DC to a low frequency like in a precision DC application, then I would use a sampling DC voltmeter to make measurements for 10 seconds and then calculate the standard deviation to get the 0.1Hz to roughly 10Hz (or bandwidth of the voltmeter) RMS noise.  Combined with a high frequency spot noise measurement, the corner frequency could then be calculated.

[npelov]

2. You didn't take the sqrt correctly. 

...  shame on me! SQRT(16kHz) is not SQRT(16)*1000! Thanks for that!

Mosfest based OPs like the MCP606 on contrast usually have a rather large 1/f noise and thus give usually poor LF noise figures. So CMOS OPs generally have a higher ration of LF noise to 1 kHz noise compared to BT based OPs. On the up side there is very little current noise for FET based OPs - so they are good for high impedance sources.

So if the input impedance is low and noise is a thing (small signal like current shunt) I better search for BJT with good 1/f noise. But if I want to measure high impedance source - like making a volt meter with 10 MOhm input impedance then current noise is more important because 10MOhm multiplies by current and adds to total voltage noise. In this case FET/MOS opamps are better choise.

The LT1006/LT1013/LT1014 are relatively low cost lower noise improved replacements for the LM358/LM324. 

LT1006 is not available locally. Buying from Farnel would cost me $4.40 in quantity of one. I wouldn't buy 10 to have them in case I need them.
LT1013 is $2.40
LT1014 is $5.60

Could you have a look at our local storage and let me know which opamps I can use for low noise, low frequency applications (not necessarily for audio). There are not that many, so it should be fast to review.

[Kleinstein]

A standard LF OP is the OP07 and newer advanved versions like OP177, OP277. For a low impedance source also the OP27 / 37 and similar are good. However they are not single supply so can't measure close to their negative supply (the LM358 and LT1013 can).

Another option can be Autozero OPs. Like the MCP6V... series (low cost  up to 5 V), AD8551, LTC2050 or old ICL7650.  The AZ OPs are especially interesting if it is really low frequencies, like < 1 Hz.

[David Hess]

The LT1006/LT1013/LT1014 are relatively low cost lower noise improved replacements for the LM358/LM324. 

LT1006 is not available locally. Buying from Farnel would cost me $4.40 in quantity of one. I wouldn't buy 10 to have them in case I need them.
LT1013 is $2.40
LT1014 is $5.60

Could you have a look at our local storage and let me know which opamps I can use for low noise, low frequency applications (not necessarily for audio). There are not that many, so it should be fast to review.

Drift is essentially very low frequency noise so if DC to low frequencies like 1Hz or 10Hz is what is important, then you want a precision operational amplifier.

The least expensive option for general purpose applications which happens to be on your list is suggested by Kleinstein, the OP07, which is a fine DC to 1Hz precision low noise amplifier specified for a 0.1 to 10 Hz noise of 0.38uV typical.  The only reason the OP07 is so cheap is that it is about the oldest of the really good low noise precision operational amplifiers and it is now widely sourced.  Something better will cost a lot more.

However the OP07 is not a single supply or rail-to-rail operational amplifier and it requires +/-3 to +/-15 volt supply voltages or a single supply voltage of 6 to 30 volts with an appropriate virtual ground.  If you need a single supply operational amplifier which will run on 3 to 5 volts, then a different and more expensive part will be required like one of the chopper amplifiers in your list.  They have a lower offset and lower drift than the OP07 but not lower noise except at even lower frequencies which is why they have a 0.01 to 1 Hz noise specification making them look better.