Are There Lower Noise IC Chips Than A TL072?

[Paul Marossy]

Are there opamps with better noise specs than a TL072? The only IC chips that I know of that might be better are the Burr-Brown OPA2134s.  Are there any others?

[StephenGiles]

How good are your ears? I have a permanent hiss in my left ear so they all sound the same to me at 9v!

[Mark Hammer]

Yes, there ARE, but their noise specs will depend upon context.  There is a comparison of noise specs in a number of "classic" op-amps in one of the issues of DEVICE I have posted at my site.  One of the things you'll note is that their ranking, as far as noise goes, will change depending upon assorted resistances at the input.

[Cliff Schecht]

Noise performance in op-amps can be a damn tricky beast and really is application specific. Some op amps that are extremely low noise in specification but won't perform so well in real life situations. Johnson noise from resistors, which occurs in EVERY resistor, gets multiplied by your op amp gain and can sometimes overwhelm your op amps low noise performance specs for example. (I read Mark's response after writing this..SORRY MARK!)

Take a look at some of your favorite op amps noise specs. They are probably all within a very close range, somewhere around 10 nV/rtHz to about 20 nV/rtHz. Sometimes, devices aren't even measured in V/rtHz but instead are measured in A/rtHz, which is the equivalent current noise from the op amp. The type of noise specified for your op amp is dependent typically on the manufacturer and type substrate the device is built on. If there are any FET's in your device, indicative of a BiCMOS or LinCMOS (or just CMOS) device, then you're going to be dealing with V/rtHz because the internal transistors are all voltage-mode devices. Bipolar processes are usually specified in pA/rtHz or fA/rtHz. Going from voltage noise to the equivalent current noise (and vice versa) is as easy as dividing and/or multiplying by the current ala Ohms law.

This is where manufacturers get a bit tricky, however. Sometimes, a device has a fairly low input current noise but a high equivalent input voltage noise. In this case they're going to specify the lower number, that being the current noise, and expect the user to know that using this device with a high impedance input will cause a lot of equivalent voltage noise. It's also important to remember that you are working in "root Hz" (the unit is "RMS Volts/Amps per root Hz"); this says that the noise decreases with a "square law" because as you go up in frequency, the noise decreases as a square (well, root) function of frequency.

As far as specific parts go, there are about 20 billion!! If you could post a circuit then I could help you find the BEST part for that circuit albeit not the cheapest. Funny enough, the OPA2134 is NOT a low noise part, or at least not as low as you think. The TL072 boasts 18nV/rtHz while the OPA134 (same as what you posted) boasts 8nV/rtHz - not what I call a low noise part! Take a look at the OPA211 (one of many just by TI) if you want a truly low noise device (link at bottom). If there is a specific company you prefer (I like Analog Devices and Burr-Brown), then I can find you more parts to look at. Again though, a specific circuit would help choose the BEST part for your application.

http://focus.ti.com/docs/prod/folders/print/opa211.html?lpos=Middle_Container&lid=Alternative_Devices

*The author takes no responsibility for any misinformation herein, I just woke up and saw this thread *

[R.G.]

The strict answer is "yes". However, simply subbing in one of them may or may not produce much change.

Like transistor gain, "lower noise" only has meaning when the circumstances and conditions are specified.

Noise situations are one of the places where picking suitable impedances (as opposed to matching impedances ) does matter. JFETs are generally good for situations where the source impedance is high. However, for low source impedances, bipolars are actually lower noise.

Bandwidth matters too. The answers get different if you talk about near-DC noise where flicker noise dominates versus the middle of the audio range where thermal noise dominates.

National Semi made some specialized opamps for low noise use in phono preamps. The LM381A is a good example. It had a typical/max noise spec of 0.5uV/0.7uV for a 10Hz-10kHz bandwidth with a 60 ohm source impedance.  This is astonishingly good. I had a friend who home-brewed a hifi preamp from one of these. He used to turn his system on, turn the preamp gain up all the way. You could hear a faint hiss in the speakers if you were right next to them. He'd then put in a vinyl disk, and flip the lever to let the needle float slowly down to the record surface. The result was like a crack of thunder when the needle touched down, absolutely deafening. The specs to look for are total noise specified for a bandwidth and source impedance, or a noise voltage specified as volts per square-root-Hz with a source impedance.

TI neatly specifies their TL072 at 4uV with a source impedance of 20 ohms and 10Hz-10kHz bandwidth. That's something like 16db more noise for similar conditions to the LM381A.

Here's the next layer of the onion:
Noise comes in two flavors, voltage noise and current noise. All active devices and resistors generate both, the two being proportional in resistors. In active devices voltage noise may dominate current noise, or the other way round. When what you're looking at is the amp itself, assuming a resistive source, you notice that (a) low source impedances (like the  60ohm and 20 ohm the opamps were measured at) will minimize current noise, leaving only voltage noise to be measured. With high source impedances (generally 10K or over) JFET input opamps tend to be quieter for similar conditions than bipolar-input types because they have almost no current input. For low input impedances, maybe 100 ohm or lower, bipolars have an advantage because their own source impedance is so low. The LM381A is a bipolar input and works well with a magnetic phono cartridge. The bipolar NE5532 is a perennial favorite in studios where its lowish input impedance helps keep 600 ohm lines quiet.

Another layer:
Everything you connect to the opamp input, like especially the feedback resistor, feeds noise into the input. So high gain with a big feedback resistor will probably also give big noise from the resistor even if you have a quietish opamp.

[Mark Hammer]

....all of which is to say that if you have a circuit where ongoing hiss would be likely to be heard (if there was any), it is a smart idea to make your boards with sockets, and have a sampler of dual op-amps of different types sitting around, and listen to what works best in that board.

[Paul Marossy]

Thanks for the replies guys!

I was actually thinking of a specific circuit when I asked the question, the GGG Digital Reverb. It's a pretty quiet circuit noise-wise, but I was just wondering if it could do even better than it could with the TL072s in it.

I get a little bit of a hiss when I have only the guitar plugged into one of my amps and the volume is cranked. When I have my pedalboard feeding the amp, and thru the reverb, it quiets down considerably. Perhaps it's an impedance thing? At least two of my pedals have a buffer built into them, and I was thinking that might be the difference.

[Cliff Schecht]

The first thing that reverb could go for is a redesign. Using IC1b to generate a stiff 4.5V reference would help a lot. This would prevent the high voltage noise that comes with a current return path with a high resistance (this is the equivalent of an unintentional resistive ground path, never good. This noise gets multiplied by the gain and you're looking at enough to cause some hissing issues. Also, that 1meg input resistor is unnecessary large and may be another culprit. Dropping the value to 100k would put the high-pass frequency response down at 16 Hz (still low enough to be inaudible) while decreasing the noise power by a factor of 10.

[Taylor]

One of the forums I hang out at is about building instruments, and "wood porn" is a common trope around those parts.

This thread is kind of "information porn", in that I get really excited reading really in-depth answers by 3 of the forum's most knowledgable people. In a way that's kind of... wrong. 

[Paul Marossy]

The first thing that reverb could go for is a redesign. Using IC1b to generate a stiff 4.5V reference would help a lot. This would prevent the high voltage noise that comes with a current return path with a high resistance (this is the equivalent of an unintentional resistive ground path, never good. This noise gets multiplied by the gain and you're looking at enough to cause some hissing issues. Also, that 1meg input resistor is unnecessary large and may be another culprit. Dropping the value to 100k would put the high-pass frequency response down at 16 Hz (still low enough to be inaudible) while decreasing the noise power by a factor of 10.

Changing the 1M input resistor to a 100K instead would be a simple thing to try.

Yeah, on IC1b.... I don't like using 1/2 of a dual opamp and grounding out the other 1/2 when you could just simply use a single opamp instead. I suppose that's another mod that could be done to the PCB.

[R.G.]

The first thing that reverb could go for is a redesign. Using IC1b to generate a stiff 4.5V reference would help a lot. This would prevent the high voltage noise that comes with a current return path with a high resistance (this is the equivalent of an unintentional resistive ground path, never good. This noise gets multiplied by the gain and you're looking at enough to cause some hissing issues. Also, that 1meg input resistor is unnecessary large and may be another culprit. Dropping the value to 100k would put the high-pass frequency response down at 16 Hz (still low enough to be inaudible) while decreasing the noise power by a factor of 10.

I think I might go a different direction.

The TL072 is a JFET input device, and has a vanishing small input current, so the input bias current through even a 1M resistor is going to be miniscule. You're only going to get the thermal noise of the 1M. The excess noise and current noise are killed by the "noiseless biasing" setup used with a resistor to a bias voltage.  There isn't much gain to speak of; IC1a has a gain of 1, IC1b has a gain of 1.25 and IC1c has a gain of 2.2, so there's not much amplification of the input noise. That means that R4-10 and R13 are going to matter almost as much as R1.

It's true that dropping R1 to 100K will still give you enough high pass to pass all the signal from a guitar, but the rolloff of the 2-4H inductance of the guitar pickup and the 100K will start dimming treble through loading at the high end of the audio range where the pickups look like 125K - 250K (computed at 10kHz).

[Paul Marossy]

The first thing that reverb could go for is a redesign. Using IC1b to generate a stiff 4.5V reference would help a lot. This would prevent the high voltage noise that comes with a current return path with a high resistance (this is the equivalent of an unintentional resistive ground path, never good. This noise gets multiplied by the gain and you're looking at enough to cause some hissing issues. Also, that 1meg input resistor is unnecessary large and may be another culprit. Dropping the value to 100k would put the high-pass frequency response down at 16 Hz (still low enough to be inaudible) while decreasing the noise power by a factor of 10.

I think I might go a different direction.

The TL072 is a JFET input device, and has a vanishing small input current, so the input bias current through even a 1M resistor is going to be miniscule. You're only going to get the thermal noise of the 1M. The excess noise and current noise are killed by the "noiseless biasing" setup used with a resistor to a bias voltage.  There isn't much gain to speak of; IC1a has a gain of 1, IC1b has a gain of 1.25 and IC1c has a gain of 2.2, so there's not much amplification of the input noise. That means that R4-10 and R13 are going to matter almost as much as R1.

It's true that dropping R1 to 100K will still give you enough high pass to pass all the signal from a guitar, but the rolloff of the 2-4H inductance of the guitar pickup and the 100K will start dimming treble through loading at the high end of the audio range where the pickups look like 125K - 250K (computed at 10kHz).

In light of this, maybe it's just due to the limitations of the Belton Digi-Log reverb unit? Hard to say without really knowing anything about what is inside of it...

[Cliff Schecht]

The first thing that reverb could go for is a redesign. Using IC1b to generate a stiff 4.5V reference would help a lot. This would prevent the high voltage noise that comes with a current return path with a high resistance (this is the equivalent of an unintentional resistive ground path, never good. This noise gets multiplied by the gain and you're looking at enough to cause some hissing issues. Also, that 1meg input resistor is unnecessary large and may be another culprit. Dropping the value to 100k would put the high-pass frequency response down at 16 Hz (still low enough to be inaudible) while decreasing the noise power by a factor of 10.

I think I might go a different direction.

The TL072 is a JFET input device, and has a vanishing small input current, so the input bias current through even a 1M resistor is going to be miniscule. You're only going to get the thermal noise of the 1M. The excess noise and current noise are killed by the "noiseless biasing" setup used with a resistor to a bias voltage.  There isn't much gain to speak of; IC1a has a gain of 1, IC1b has a gain of 1.25 and IC1c has a gain of 2.2, so there's not much amplification of the input noise. That means that R4-10 and R13 are going to matter almost as much as R1.

It's true that dropping R1 to 100K will still give you enough high pass to pass all the signal from a guitar, but the rolloff of the 2-4H inductance of the guitar pickup and the 100K will start dimming treble through loading at the high end of the audio range where the pickups look like 125K - 250K (computed at 10kHz).

I'd love to measure how "noiseless" that biasing configuration really is, especially when the same relatively high impedance 4.5V source is being used in multiple places. Granted nothing on the 4.5V draws any appreciable current (like R.G. was hinting at, the TL0xx series draw pA of current), but I would still think that there is potential for crosstalk between the wet and dry stages. Turn down the mix all the way (completely dry) and crank up a LOUD amp, this will quickly let you know if you are getting bleedthrough between each stage .

I really just hate seeing a TL072 being wasted when it would make a great 4.5V low impedance reference. Also, you're correct about the guitar pickups being loaded down easily. That's something that I usually forget to take into effect because everything in the synth world is buffered (including most of my guitars!).

[R.G.]

I'd love to measure how "noiseless" that biasing configuration really is, especially when the same relatively high impedance 4.5V source is being used in multiple places. Granted nothing on the 4.5V draws any appreciable current (like R.G. was hinting at, the TL0xx series draw pA of current), but I would still think that there is potential for crosstalk between the wet and dry stages. Turn down the mix all the way (completely dry) and crank up a LOUD amp, this will quickly let you know if you are getting bleedthrough between each stage .

The impedance of the biasing stage is really the impedance of that 47uF bypass cap to ground. That's about 41 ohms at 82Hz. IC2B inverting input is the only place with a measurable amount of current going it, that being the signal current through that 39K plus whatever (unknown) is coming through R13. Any signal from R7 is attenuated by 41/(39000+41) = -59.6 db. Not perfect, but not bad either. That's the worst point until the high frequency effects start running the impedance up on the bypass cap, so it only gets better, possibly into the low ultrasonic, depending of course on what the quality of the cap is.

I did a quick look for the textbook that introduced me to the "noiseless biasing" term. It may actually be in Ott's Noise Reduction Techniques in Electronic Systems
but I didn't find my copy on the first pass.

I really just hate seeing a TL072 being wasted when it would make a great 4.5V low impedance reference. Also, you're correct about the guitar pickups being loaded down easily. That's something that I usually forget to take into effect because everything in the synth world is buffered (including most of my guitars!).

You're right about that; I like buffered references too. But buffered references are not necessarily perfect. I've run into situations where a buffered reference can actually pick up noise and then buffer it out to other places. It's really important in buffered references to keep noise under control and keep gain in the buffer low; no amplifying a Vbe up to 50V! 

If I remember correctly, the name actually came from its use in bipolar biasing instead of the two-resistor base bias of the "stabilized bias circuit". Using one more resistor and one more cap let you get independent from the noise of the power supply and the two biasing resistors as well as the significant current noise in them, subbing in the thermal noise of one resistor, which is about as low as you can get a bias "voltage" and still have any input impedance at all.

[Paul Marossy]

I think I might have to retract my statement about the noise possibly being generated at the Digi-Log unit. The noise specs appear to be pretty good to me. http://www.smallbearelec.com/Projects/BTDR-1H.pdf

Maybe it just has to do with my tube amp more than the reverb circuit? This is the schematic for the amp that I am using with this reverb circuit: http://www.diyguitarist.com/PDF_Files/MarkXSchemR1.pdf

Like I mentioned earlier, it only hisses noticeably when I have a guitar plugged straight into the amp (thru the reverb circuit) with the volume up about 1/2 way. When I use my normal pedal board setup (with two commercial, buffered pedals on it), the hiss isn't really there anymore.

[R.G.]

...
Maybe it just has to do with my tube amp more than the reverb circuit? This is the schematic for the amp that I am using with this reverb circuit: http://www.diyguitarist.com/PDF_Files/MarkXSchemR1.pdf

Like I mentioned earlier, it only hisses noticeably when I have a guitar plugged straight into the amp (thru the reverb circuit) with the volume up about 1/2 way. When I use my normal pedal board setup (with two commercial, buffered pedals on it), the hiss isn't really there anymore.

I have messed with a few amps and there are some things that make me nervous there. The first stage being grid-leak biased with a 10M resistor is one of them. Grid leak biasing produces high gain, and since thermal noise goes up with resistance, a 10M resistor is generating a lot more noise than the more traditional 1M value. What happens to the sound if you bag the cap and 10M and simply use the 1M direct coupled?

[sean k]

Um, I've always been pleased with the results of the 5534 singles and the 5532 dual opamps where noise might be an issue. Nice and cheap too.

[R O Tiree]

R.G. - Didn't I read somewhere on your site about using an LM386 with its inputs tied together providing an output floating perfectly at Vcc/2 ? How would that figure in the noise stakes?

[Paul Marossy]

...
Maybe it just has to do with my tube amp more than the reverb circuit? This is the schematic for the amp that I am using with this reverb circuit: http://www.diyguitarist.com/PDF_Files/MarkXSchemR1.pdf

Like I mentioned earlier, it only hisses noticeably when I have a guitar plugged straight into the amp (thru the reverb circuit) with the volume up about 1/2 way. When I use my normal pedal board setup (with two commercial, buffered pedals on it), the hiss isn't really there anymore.

I have messed with a few amps and there are some things that make me nervous there. The first stage being grid-leak biased with a 10M resistor is one of them. Grid leak biasing produces high gain, and since thermal noise goes up with resistance, a 10M resistor is generating a lot more noise than the more traditional 1M value. What happens to the sound if you bag the cap and 10M and simply use the 1M direct coupled?

You mean delete the 0.1uF input cap and the 10M resistor to ground and use a 1M series resistor instead? I guess I could try that and see what happens...

[Cliff Schecht]

Um, I've always been pleased with the results of the 5534 singles and the 5532 dual opamps where noise might be an issue. Nice and cheap too.

You and PAiA both. We use them in a lot of our mic pre's and a lot of other places and they give damn good results for the price IMO. Obviously not the best, but damn suitable for most non-precision guitar/voice applications.