signal to noise ratio


by m718
Tags: noise, ratio, signal
m718
m718 is offline
#1
Aug4-09, 07:45 PM
P: 88
What are some ways to improve signal to noise ratio for a microphone amplifier, beyond getting low noise components. My amplifier is a single opamp.
Phys.Org News Partner Engineering news on Phys.org
Fiber-optic microscope will help physicians detect cancer, diseases at early stages
Wind tunnel tests support improved aerodynamic design of B61-12 bomb
Smart sensor technology to combat indoor air pollution
chroot
chroot is offline
#2
Aug4-09, 08:03 PM
Emeritus
Sci Advisor
PF Gold
chroot's Avatar
P: 10,424
Increase the amplitude of the signal (assuming the noise stays constant, as it does in most circuits).

- Warren
berkeman
berkeman is offline
#3
Aug4-09, 08:05 PM
Mentor
berkeman's Avatar
P: 39,578
Quote Quote by m718 View Post
What are some ways to improve signal to noise ratio for a microphone amplifier, beyond getting low noise components. My amplifier is a single opamp.
What are your thoughts? Do you have a particular problem with noise now, or are you just planning the circuit, and want to make it the best it can be given some constraints?

Is the microphone preamp right at the microphone? Or is the microphone passive, and the first amplifier is a cable length away? What kind of noise do you anticipate? Just thermal noise in the circuitry, or 50/60Hz hum too?

Have you looked into the general concepts of low-noise circuit design techniques and shielding? (Was there a recent thread here about that? ... I need to look now....)

m718
m718 is offline
#4
Aug4-09, 08:28 PM
P: 88

signal to noise ratio


Quote Quote by berkeman View Post
What are your thoughts? Do you have a particular problem with noise now, or are you just planning the circuit, and want to make it the best it can be given some constraints?

Is the microphone preamp right at the microphone? Or is the microphone passive, and the first amplifier is a cable length away? What kind of noise do you anticipate? Just thermal noise in the circuitry, or 50/60Hz hum too?

Have you looked into the general concepts of low-noise circuit design techniques and shielding? (Was there a recent thread here about that? ... I need to look now....)

Its a dynamic mic on the same board as the amp(1"x1") I have a passive notch for the 60hz shielded( the amp is in a metal box) so the noise thats left is thermal the gain of the opamp is set to 1500 and I need the signals down to 10 nanovolts even 1nV if if its won't be too expensive but around 200nV is what it can recieve right now.
Mike_In_Plano
Mike_In_Plano is offline
#5
Aug4-09, 10:27 PM
P: 557
Hello, Mike here,

First, ascertain the impedance of your source, this will dictate the type of amplifier you wish to use. Each amplifier is defined by an equivalent input noise voltage, e_n, and an equivalent input noise current, i_n. There's also a low frequency noise, 1/f noise. I haven't worked much with this lower frequency noise because I seldom design anything of great sensitivity below 100Hz, so I can't help you with it.

In any case, I can help you with en, in, and selecting a good amp.

Generally, lower noise op amps will have an e_n of .9 to about 10 nanovolts per sqrt(Hz). The current noise varies more greatly, with an i_n of 1 to 10,000 nano amps per sqrt(Hz).

Here's where your input impedance comes to play, i_n is being created by the op amp and is flowing through your mic to create a noise voltage across your mic. So,

v_noise_mic = z_mic x i_n

The resulting input noise will then be:

v_noise_input = sqrt (v_noise_mic^2 + e_n^2)

Ideally, you'd find the op amp that keeps both of these numbers to a minimum and approximately the same. For low Z microphones, with an output of 500 ohms or less, there are specialized bipolar amps with extra large input transistors. Linear Technology has a wide selection of these, some with en less than 1 nv / rtHz.

For mid z microphones, with an output of 1-5 k ohms, bipolar amps are still appropriate, but the ones with en < 2 nV/rtHz usually have too much i_n, so you have to dig around for an en less than 5 nv/rtHz and an i_n less than 2pA/rt Hz.

For high z microphones, with an output above 10 k ohms, it's usually best to go with a low noise FET amp. FET amps have negligible current noise, but suffer from a step up in voltage noise. e_n from 6 to 12 nv / rtHz is common. Curiously, FET amps give you even better noise performance as your impedance continues to increase, because the mic's output voltage is increasing and the FET amp still has negligible noise current.

FET input amps can also be ganged together in parallel to give you better noise performance. The reason is that each one is an uncorrelated noise source, the noise each is making is off doing it's own thing and isn't additive. The equation is:

e_n / sqrt (N), where N is the number of parallel amps.

Hence, four cheap amps in parallel will halve the noise voltage and quarter the noise power! Just remember that each must have it's own feedback circuit, etc, and the outputs must be tied together through resistors, otherwise, they'll fight.

Some good parts are:

For up to 1k ohm, LT1128
x
For over 10k ohm, OPA604
m718
m718 is offline
#6
Aug4-09, 10:44 PM
P: 88
Quote Quote by Mike_In_Plano View Post
Hello, Mike here,

First, ascertain the impedance of your source, this will dictate the type of amplifier you wish to use. Each amplifier is defined by an equivalent input noise voltage, e_n, and an equivalent input noise current, i_n. There's also a low frequency noise, 1/f noise. I haven't worked much with this lower frequency noise because I seldom design anything of great sensitivity below 100Hz, so I can't help you with it.

In any case, I can help you with en, in, and selecting a good amp.

Generally, lower noise op amps will have an e_n of .9 to about 10 nanovolts per sqrt(Hz). The current noise varies more greatly, with an i_n of 1 to 10,000 nano amps per sqrt(Hz).

Here's where your input impedance comes to play, i_n is being created by the op amp and is flowing through your mic to create a noise voltage across your mic. So,

v_noise_mic = z_mic x i_n

The resulting input noise will then be:

v_noise_input = sqrt (v_noise_mic^2 + e_n^2)

Ideally, you'd find the op amp that keeps both of these numbers to a minimum and approximately the same. For low Z microphones, with an output of 500 ohms or less, there are specialized bipolar amps with extra large input transistors. Linear Technology has a wide selection of these, some with en less than 1 nv / rtHz.

For mid z microphones, with an output of 1-5 k ohms, bipolar amps are still appropriate, but the ones with en < 2 nV/rtHz usually have too much i_n, so you have to dig around for an en less than 5 nv/rtHz and an i_n less than 2pA/rt Hz.

For high z microphones, with an output above 10 k ohms, it's usually best to go with a low noise FET amp. FET amps have negligible current noise, but suffer from a step up in voltage noise. e_n from 6 to 12 nv / rtHz is common. Curiously, FET amps give you even better noise performance as your impedance continues to increase, because the mic's output voltage is increasing and the FET amp still has negligible noise current.

FET input amps can also be ganged together in parallel to give you better noise performance. The reason is that each one is an uncorrelated noise source, the noise each is making is off doing it's own thing and isn't additive. The equation is:

e_n / sqrt (N), where N is the number of parallel amps.

Hence, four cheap amps in parallel will halve the noise voltage and quarter the noise power! Just remember that each must have it's own feedback circuit, etc, and the outputs must be tied together through resistors, otherwise, they'll fight.

Some good parts are:

For up to 1k ohm, LT1128
x
For over 10k ohm, OPA604

I'm using OP-27G input noise voltage 3.8 nV at 1khz. I don't know how to measure the impedance of the mic but the resistance with a multimeter is 1KOhm.
For parallel would I just put a few more opamps in the same way I have the current one and just connect the signal to all and connect outputs together?

I have signal going to +, 500ohm going from + to ground, 500ohm from - to ground, and 750kohm from out to -.

frequency is 300hz to 5000hz
m718
m718 is offline
#7
Aug4-09, 10:51 PM
P: 88
I heard AM radio used a system where they add many sine waves together to lower the noise
it only works for fixed frequencies but thats not a problem I read here:
http://www.4p8.com/eric.brasseur/receiv.html but I don't really understand what they mean.
Mike_In_Plano
Mike_In_Plano is offline
#8
Aug4-09, 10:57 PM
P: 557
I don't know about the AM radio thing. I'm doubtful that anything other than good engineering practice will get you out of the noise situation.

I accidentally posted the last message before I finished. Just a couple more notes:

I'd reduce the gain on that first stage. Why? Because the op amp isn't a perfect device, particularly as you go up in frequency. With a gain of 1500, you're probably loosing gain and delivering distortion in your upper frequencies simply because the amp doesn't have enough open loop gain.

For example, suppose your amp had an open loop gain of 2 million and a gain bandwidth of 10MHz. That means that the gain is 2 million only for a DC signal at 10MHz, the gain is 1. For every decade of frequency you drop, the gain goes up by a factor of 10, so you have:
gain = 10 @ 1MHz
gain = 100 @ 100kHz
gain = 1000 @ 10kHz Ooops, people can hear 10kHz, and we don't have enough gain!

A gain of 100 would be far more reasonable for a single stage - assuming you have a gain bandwidth of at least 10MHz.

Also, if you're oscillating out of the audible band, that can make noise in-band.


Some good parts are:

For up to 1k ohm, LT1128
For up to 15k ohm, LT1007 and LT1677
For over 15k ohm, OPA604 (my favorite!)

Or, if your super cheap, TL084, with 4 devices paralleled for about 9nV/rtHz.

In all these cases though, a well designed discrete transistor amp will run circles around an op amp, but that's another story.

- Mike
Mike_In_Plano
Mike_In_Plano is offline
#9
Aug4-09, 11:14 PM
P: 557
Gotcha,

I read the radio tutorial. What they're saying is that the noise is random, but the sine wave you're looking for isn't. If you can keep adding the sine waves together, the result is an ever increasing sine wave, but since the noise signals are random, they don't accumulate. So with enough averaging, you can see the sine wave and eventually remove the noise. The S/N increases like this:

V_sine/V_noise (averaged) = V_sine/V_noise (typical) x 1/sqrt(N)

Where N is the number of times you average.

This is all well and cool if you know where the sign wave begins and ends, but if you don't the result of one addition is just as likely to cancel out the last sine wave as it is to help. Still, the technique has some cool uses. For example, ever wonder how a cruddy antenna within a GPS system is able to pick up a small satellite transmitter from over 10,000 miles away? It's the same reason that the GPS receiver takes so long to lock. It attempts to average a known signal over and over again until it just happens to get it right. When the receiver "guesses" the phase of the signal, the averages add up to something it can see - like the sine wave coming out of the noise.
m718
m718 is offline
#10
Aug4-09, 11:52 PM
P: 88
Quote Quote by Mike_In_Plano View Post
I don't know about the AM radio thing. I'm doubtful that anything other than good engineering practice will get you out of the noise situation.

I accidentally posted the last message before I finished. Just a couple more notes:

I'd reduce the gain on that first stage. Why? Because the op amp isn't a perfect device, particularly as you go up in frequency. With a gain of 1500, you're probably loosing gain and delivering distortion in your upper frequencies simply because the amp doesn't have enough open loop gain.

For example, suppose your amp had an open loop gain of 2 million and a gain bandwidth of 10MHz. That means that the gain is 2 million only for a DC signal at 10MHz, the gain is 1. For every decade of frequency you drop, the gain goes up by a factor of 10, so you have:
gain = 10 @ 1MHz
gain = 100 @ 100kHz
gain = 1000 @ 10kHz Ooops, people can hear 10kHz, and we don't have enough gain!

A gain of 100 would be far more reasonable for a single stage - assuming you have a gain bandwidth of at least 10MHz.

Also, if you're oscillating out of the audible band, that can make noise in-band.


Some good parts are:

For up to 1k ohm, LT1128
For up to 15k ohm, LT1007 and LT1677
For over 15k ohm, OPA604 (my favorite!)

Or, if your super cheap, TL084, with 4 devices paralleled for about 9nV/rtHz.

In all these cases though, a well designed discrete transistor amp will run circles around an op amp, but that's another story.

- Mike
I found some preamp schematics but they don't work when I put them together do they have to be connected to a power amplifier if so can I use one opamp as the amplifier?
Mike_In_Plano
Mike_In_Plano is offline
#11
Aug5-09, 02:17 PM
P: 557
Good choice in op amps, the OP27 should perform very well up to about 7K. At that point, it's current noise, i_n, will create about 3.5 nv across 7K and be about equal to the op amp's internal noise.
.
You're microphone's impedance probably varies a bit with frequency. You probably want to pick a test frequency where noise is most annoying, around 3 kHz, to check you're mic's impedance. Here's what you need:
.
1. A 3kHz signal source - about 50mV
2. A scope or sensitive AC meter (good for measuring up to 3 kHz and down to 25mV accurately)
3. An assortment of resistors, or a pot, spanning 1k-10K.
.
As a heads up, always connect your equipment to the same outlet or power strip when making low-noise measurements - it keeps the pops and clicks from other equipment from bothering you as much.
.
Procedure:
1. Make a common ground point where the following meet:
. a. The scope ground (or meter ground)
. b. The signal generator's ground
. c. One terminal of the resistor
.
2. Make a measuring point where these meet:
. a. The scope probe (or meter input)
. b. The other terminal of the resistor
. c. One terminal of the mic
.
3. Finally, make a point where the test signal is injected:
. a. The other terminal of the mic
. b. The output of the signal generator.
.
4. Adjust the signal generator for 50mV out at 3 kHz ( you may want to borrow the scope for this).
.
5. Check the level of the 3kHz signal at the measurement point from 2.
.
6. Change out the resistor until the measurement seen in step 5 is roughly 25 mV.
.
7. The resistor value that gives you about 25 mv is the approximate impedance of the mic.
.
As for the circuit, the first thing I'd do is ditch the 500 ohm resistor across the mic. It's costing you signal level and it's probably interacting with the mic to give you a crazy frequency response. Instead, I'd put a small value NPO cap in series with a resistor, about 100pF and 100 ohms, from that point to ground. That's probably not the best loading network, but it's the most likely one to keep you from oscillating.

Next, I'd look at the open loop gain vs. frequency curve. It shows about 63 dB at about 5kc.

That's A_v = 10^(63 / 20), so A_v = 1400

So, decrease your closed loop gain, so that the op amp can work properly. The data sheet indicates that the amp is very linear, so it would likely have good distortion with a closed loop gain as high as 100-140.

I'd toss in a 49.9K metal film feed back resistor (out to -in) with a 499 ohm metal film resistor from -in to ground, and see how that treats your noise.

You may want to toss in a power supply decoupling filter as well. While op amps can reject DC changes in the power supply, they are much worst at rejecting AC changes and the problem becomes worse as the frequency increases. Distortion or even oscillation may occur if upstream stages shake the power supply and it gets back to your pre amp. Typically, I toss in a couple of 100 ohm resistors in series with the power supplies. Then either bypass them to ground with 1 uF ceramics (rate them for 2x or more the supply voltage), or 10-22uF (rate for 1.25x or more the supply voltge) electrolytics.

As for the parallel amps, they're essentually seperate amps, say four, each it's +in input tied to the mic and each with it's output going through a 100ohm (or more) resistor to a common output. With what I know about your amp and mic, I wouldn't do this because the i_n of your amp is fairly high and the mic's impedance is likely to be high as well.
.
. Best Wishes,
. - Mike


Register to reply

Related Discussions
matlab-compute the distribution of the noise signal Math & Science Software 1
signal/noise in Raman spec Chemistry 1
signal, noise, and, unequal variances Calculus & Beyond Homework 0
What is meant by "signal-to-noise ratio?" General Engineering 15
Poisson stats: signal to noise Introductory Physics Homework 3