Noise Figure and Dynamic Range??

[MatthewF]

Hi there, I'm hoping that someone can help me clarify some points with regards to noise and interfacing, as I'm starting to get a little confused (maybe trying to think too hard!)

I was recently reading the data sheet for an INA134 instrumentation amp and considering using it to realise a balanced line-level input. I noticed that it has a noise voltage of 52nV/rootHz (7.35uV with 20kHz bandwidth).

Assuming a 200ohm source impedance (about 1.8nV/rootHz) this would result in a noise figure of 29.5dB.

This sounds terrible, but I assume the INA134 is acceptable for a line level input due to the higher signal level? If this is the case, does that imply that the calculation of a noise figure is only relevant to high gain mic amps etc. (where input signal level is very low)?

This leads me to ask what measurement is suitable for gauging the noise of a line level input. It is clearly relative to signal level. I've seen spec sheets citing THD+Noise as a percentage. Does this refer to the percentage the of total power which is contained within noise and distortion, at a given signal level?

If this given signal level were +4dBu, that would imply a signal voltage of 1.23vRMS. How would this 1.23v signal level and 7.35uV voltage noise be compiled into a useful parameter with which to compare designs?

Thanks for any help,

Matthew

 

[bcarso]

[quote author="MatthewF"]---snip---This sounds terrible, but I assume the INA134 is acceptable for a line level input due to the higher signal level? If this is the case, does that imply that the calculation of a noise figure is only relevant to high gain mic amps etc. (where input signal level is very low)? --snip---

If this given signal level were +4dBu, that would imply a signal voltage of 1.23vRMS. How would this 1.23v signal level and 7.35uV voltage noise be compiled into a useful parameter with which to compare designs?

[/quote]

You are on it. It's just a matter of what S/N ratio you are after, and in many cases how many other noise sources are already dominating the net result. There's no qualitative difference due to signal level. But clearly it will be easier to get good S/N when the reference signal is a large one rather than a small. In the case of your 1.23V rms signal and 7.35uV rms noise the S/N ratio will be 104.5dB, a good but not great result.

Bear in mind that there will always be a current noise contribution, which will produce another noise voltage flowing in your source impedance. There is usually a small correlation of this noise signal with the voltage noise which is typically neglected, and indeed is usually not very important at audio frequencies.

If it weren't for the current noise we could just keep paralleling devices to reduce the voltage noise.

Line amps are typically looking at low source impedances---sometimes as low as a few ohms or less, other times higher due to a final resistor in series, or an L-R combo, to lessen sensitivity to capacitive loading that might provoke oscillation.

If you had a source that somehow managed itself to have amazingly high signal-to-noise ratio, it might warrant a line amp with comparable or better S/N. 24 bit audio is usually really 18-20 bit audio with noise bits, but they were all real this would imply a S/N ratio of >144dB. Provided the output voltage range is not too restricted this is achievable, if not easy. It might be fun to do, but it's mostly pointless. Maybe good for marketing :razz:

 

[PRR]

> calculation of a noise figure is only relevant to high gain mic amps etc. (where input signal level is very low)?

Noise Figure is relative to an ideal resistor.

The hidden assumption is that many transducers (microphones) act a whole lot like resistors.

A 200Ω dynamic mike in a vacuum will output a noise voltage nearly the same as a perfect 200Ω resistor. (Put it in dead-quiet air, and the thermal noise of the air adds to the electrical noise; put it in about any real studio and the truck and blower rumble adds to the theoretical noise sources.)

Note that the now-popular Condenser mikes have a very different noise picture. In fact they are a lot like preamps.

Take that dynamic mike. At most musical levels, the output is inconveniently low. A complete recording system has many stages. It is bad enough to need one very-low-noise stage; we don't want to have to make ALL our stages that expensive.

So the first thing we do is bump it up in level. That raises the signal, it also raises the noise. Ideally the S/N ratio is the same as, or not much worse than, the S/N in the room.

Signal of a 200Ω dynamic is about 10mV; noise us about 0.2uV. "About 1V" is the widely accepted (with detail differences) "line" level. So we boost-up the dynamic's 10mV by a factor of 100 or 40dB to get 1V. The noise is also increased to about 20uV.

Now if we feed this signal into your 8uV input noise amp, we have 20uV of source noise and 8uV of added noise. They are not correlated so they don't sum to 28uV, more like 21.5uV. This is a 0.65dB reduction in S/N. This is not much worse than using an ideal noiseless input stage, and it was done with $0.002 transistors and large protective resistors, instead of big fat unprotected transistors or elaborate transformers.

The noise seems to be mostly the resistors. If 200Ω makes 0.2uV, then 20,000Ω makes 2uV. The noise-gain of that topology is 2, and there are two sides. I'm not sure the resistor noise alone sums to 7uV, but it is most of the output noise. Using the 3-amp topology would lose these series resistors and lower the noise level, but the resistors are valuable in preventing "accidents" and, as we saw, do not degrade S/N much for practical situations.

 

[SSLtech]

[quote author="bcarso"]...but they were all real this would imply a S/N ratio of >144dB... <snip> ...It might be fun to do, but it's mostly pointless. Maybe good for marketing :razz:[/quote]
I believe that Rupert Himself (doff cap, touch forelock, bow, scrape) has an amplifier on the bench that surpasses even these figures... What he does with it and how much it costs... and then how nice it actually sounds are all possibly unrelated aspects however!

:green:

Keith

 

[CJ]

There are probably a lot of AES papers on noise vs headroom etc, but I do not have an index. Does anyone know how or where to find a listing of AES papers? Rein Narma's first AES paper was on Noise vs Distortion I believe, but I do not think it is in there. Thanks and sorry of OT.

cj

 

[bcarso]

[quote author="SSLtech"][quote author="bcarso"]...but they were all real this would imply a S/N ratio of >144dB... <snip> ...It might be fun to do, but it's mostly pointless. Maybe good for marketing :razz:[/quote]
I believe that Rupert Himself (doff cap, touch forelock, bow, scrape) has an amplifier on the bench that surpasses even these figures... What he does with it and how much it costs... and then how nice it actually sounds are all possibly unrelated aspects however!

:green:

Keith[/quote]

There's quite a bit of room left beyond 144 dB.

If the source Z is arbitrarily small we can parallel stuff 'til the cows come home. If the output swing is arbitrarily large we can improve dynamic range above noise arbitrarily. If the input referred noise is 10nV rms (conceivable with a bunch of paralleled devices) and the output swing is 100V rms (also conceivable---although you would have to ask what we would ever DO with such an output) then we could have a dynamic range of 200 dB.

Cooled FETs can have awfully low current noise at low frequencies, but eventually induced gate noise takes over---the thermal channel noise capacitively coupling into the gate. Thermal channel noise can be reduced a bit by cooling, but for silicon FETs carrier freezeout stops improvement below about 100 Kelvin.

 

[CJ]

You can cancel stuff too.

 

[MatthewF]

Thanks very much for all your replies, they've certainly helped to straighten this out in my head!

It seems that using this IC for a +4dBu input stage would be fine. The 104.5dB S/N ratio that bcarso quoted is certianly fine as my soundcard quotes a 105dB S/N ratio.

It seems, however, that it might be a bit noisy if a -10dBV unbalanced signal were fed into it. If -10dBV is taken as 0.32vRMS then the resulting S/N ratio would be just 93dB.

Does this imply that the INA134 is only good for 'pro' level interfacing?

If I needed an input quiet enough to deal with -10dBV signals would I be better off with a lower noise amp e.g. NE5534, configured with lower value resistors, or maybe a 3 amp configuration as PRR suggested?

Thanks again for any help with this,

Matthew

 

[cuelist]

[quote author="MatthewF"]If I needed an input quiet enough to deal with -10dBV signals would I be better off with a lower noise amp e.g. NE5534, configured with lower value resistors, or maybe a 3 amp configuration as PRR suggested?[/quote]

It's not like -10dBv is that low a level.

Why don't you consider Ted Fletcher's ingenious "superbal" circuit, it only uses two opamps and has equal impedance to ground.

 

[TedF]

I was just going to come in on this one anyway!..... but not very seriously.

It was as a note to PRR:-

If you have a ribbon mic with good chunky transformers, then the DC resistances in the system are almost zero; and as there is no resistance, then isn't it a fact that there's no noise generators? :shock:

So SURELY everyone who wants low noise recordings should be using ribbons!

 

[gyraf]

When the DC resistances in the system are almost zero, the DC-noise will be close to zero as well. But sadly, up in the audio band, it's the audio band impedances that are adding noises..

 

[TedF]

Very true; but there was a serious side to the joke; if the DC resistances are kept extremely low, and the self-noise of the mic amp is also very low, then it's not that difficult to beat the '128dB below input' that we are all taught is the limit of input noise..... it's all a question of terminology and specifications. (we are force-fed the rule that all mic inputs are 200 ohm).
But let's not get too serious! :?

 

[gyraf]

:razz:

 

[Rochey]

[quote author="SSLtech"][quote author="bcarso"]...but they were all real this would imply a S/N ratio of >144dB... <snip> ...It might be fun to do, but it's mostly pointless. Maybe good for marketing :razz:[/quote]
I believe that Rupert Himself (doff cap, touch forelock, bow, scrape) has an amplifier on the bench that surpasses even these figures...
Keith[/quote]

What he doesn't tell you is the voltage rails this beastie runs from... I imagine that that having a voltage swign of around 25V kinda dwarfs any internal noise.

 

[PRR]

> it might be a bit noisy if a -10dBV unbalanced signal were fed into it. If -10dBV is taken as 0.32vRMS then the resulting S/N ratio would be just 93dB.

"-10dB" is not 0.3V, but 2V.

The traditional "+4dB" and "-10dB" specs are taken on traditional VU meters displaying speech/music. That was all we had in 1939, and well into the 1970s for US gear. Peaks will be 14dB to 20dB higher; 16dB is a good ballpark. That's why a "+4dBm" output has its distortion rated at +20dBm. "-10dBV" happened in part because early chips would slew-limit above about 2.8V peak or a 2V RMS sine.

So a 16-bit source, such as a CD player, scaled for Hi-Fi levels (which have been essentially -10dBV since the second generation of Dolby chips forced some standardization of Hi-Fi levels), will output 2V RMS for a maximum unclipped sine. The noise level is (ideally) 96dB lower: 31.7 microVolts. Compared with your 8uV input, source noise overwhelms input noise. It is indeed suitable for 16-bit "-10dBV" sources, or 17-bit sources, and only modestly embarrassed by an 18-bit source.

A nominal 24-bit source has a range of 144dB. If we assume that we can really get 24-bit ADCs and DACs, and that we switch from simple robust stages like INA134 to dedicated stages with 1 microVolt noise, we need peak levels equivalent to a 15.85V RMS sine wave or 22.4V peak. Given the abundance of range, we would probably take a 20dB headroom allowance above VU meter speech/music indication. That puts nominal VU level at 1.6V. This is so close to the conventional "+4dBu" 1.23V levels that we may as well use that, rather than re-invent the entire studio level diagram. Still assuming 20dB headroom above nominal level, we have max output of 12.3V RMS sine, 17.4V peak. Then to get noise 144dB below 12V RMS, we need to get noise down to 0.76 microvolts.

A dumb old NE5534 will get in that zone, if you keep the resistors below a few Kohms.

Note, however, that we can't take any gain or loss without hitting a limit. If the input can really be as high as 17V peak and the chip can do 20V peak, it will clip with 1.4dB of gain. If the chip noise is 0.5uV and the digi-noise is 0.76dB, we can only reduce level 3.6dB before digi-noise is masked by input noise (that's simplistic: digi-noise and random noise don't sound the same to the ear; let's pretend.)

Frankly, the only reason to have "24-bit" is to eliminate the Gain knob. 20 or 22 bits will encode any practical electronic source without analog level set, and then you can normalize in the CPU. Since computers have 8 fingers, we give them a 24-bit DAC/ADC, but it takes heroic work just to get bit 1 smaller than bit 2, since both are so tiny and near the limits of analog noise and practical digital switchery.

> it might be a bit noisy ... S/N ratio would be just 93dB.

"Just 93dB". Oh you kids.

A lot of good work was done on tape. We quoted S/N re: VU nominal, about 55dB or 60dB. We ran peak inputs up to 20dB hotter but they came back only 14dB hotter: say 16dB peak factor. 71dB or 76 dB total dynamic range. Only the best vinyl beat this; most was worse. And it was good, even for full orchestra with silent pauses.

Home listening systems rarely beat 90dB range from room noise to peak amp/speaker effort, and are more often run near 70dB dynamic range. Already we lost the bottom 1 to 4 bits of a 16-bit signal.

Popular music has not exceeded 25dB loud/soft ratio since it moved into crowded clubs and onto 78s. Actually, a harpsichord is only 25dB above the background noise of all but the quietest rooms. Because tape/vinyl noise is distinct from room noise, and digi-noise is garbage, we like to have much more range on the media than in the room. Still, you can master most pop and a lot of small ensembles on 50dB gear if you set levels carefully.

 

[PRR]

> If you have a ribbon mic with good chunky transformers, then the DC resistances in the system are almost zero

Make the iron as chunky as you want. In any good ribbon mike, the transformer resistance is very-small compared to ribbon resistance. Ribbon resistance is our rock/hard-place. We can reduce it with a short thick ribbon, but the reduced length and increased mass reduces output (remember we don't care about Noise but about Signal/Noise).

In fact what you do is design ribbon electrical resistance, multiplied by magnetic coupling and ribbon area, equal to the impedance of air at the top of the audio band. And then because that gives a droopy top, you dance-in some horn-loading or resonators.

For practical audio microphones, ribbons, magnets, the electrical impedance and output voltage comes out very low. We don't mind, because we can design a low-loss transformer to get to a more practical impedance, and IF our ribbon couples enough to make adequate power, to a practical voltage.

But resistance is unavoidable in practical copper/aluminum dynamic motors/transducers. (Superconducting motors change the landscape; but for speakers, ribbons, and dynamics all they do is better expose acoustic resistance, which we can usually ignore for practical designs.)

Dynamic transducers, scaled (or transformered) to different resistance, give output voltage proportional to square-root of resistance.

If a 10-turn dynamic-mike voice-coil, in front of a loud Fender, gives 1V at 10Ω, then you would think a 20-turn coil would give 2V at 20Ω. But twice as many turns won't fit in the magnet gap,a nd the increased mass reduces output. You have to switch to a wire of half the cross-section, so you really get 40Ω. Compare the maximum available power: 1^2*10Ω= 0.1 Watts, 2^2*40Ω= 0.1 Watts. No difference. POWER conversion efficiency is not improved by changing resistance.


> as there is no resistance, then isn't it a fact that there's no noise generators?

Ha-ha, but many folks here don't have the background to "get" the joke.

ALL perfect resistors at given temperature have the same noise POWER.

If resistance is very-very low, noise voltage is low, yes, but noise Current is very-very high.

Noise Voltage will scale as square-root of resistance.

Since we saw that scaling resistance changes signal Voltage as square-root of resistance, the Signal/Noise ratio is not changed by changing resistance.

So we pick resistance by other criteria. Practical ribbons wind up near 0.1 to 1.0 ohms. Practical voice coils wind up in the 2 to 200 ohm range.

For short lines, the line impedance could be the same as the transducer resistance. But we know from the common 8Ω loudspeaker lines that you can just about get across the room before line resistance starts reducing output, forcing fatter costlier wires. For minimum wire cost, considering that plastic is cheaper than copper, we might try very high impedance. But we know from 50KΩ hi-Z mikes and guitar cords that beyond a dozen feet, cable capacitance sucks-off the highs. For practical wire construction, a "good" impedance for a long line is 20Ω to 500Ω. We have 37Ω, 600Ω, and other standards: dual-winding transformers offer a 4:1 choice which used to be 500Ω:125Ω but evolved to 600:150. Dynamic mike makers quoting Voltage output like to cheat to a higher impedance; 200Ω has become the accepted nominal pro mike impedance. You can run 200Ω quite a long way on light-gauge cable without great increase of series resistance or loss of treble.

> the '128dB below input' that we are all taught is the limit of input noise.....

You are being sloppy. It is ~128dB below 1dBm for audio-band and room temperature. We'll take bandwidth and temperature as pretty constant; only radio-astronomers can soak their sources in liquid Helium, musicians hate that; bandwidth of more than 20KHz is not audible noise and less than 10KHz will sound dull even to my old ears.

Since the -128dB below one milliWatt is a POWER specification, we can transform to any impedance and get a different ratio of voltage and current. Self-noise of a 200Ω dynamic mike is 0.2uV and 1nA; a 0.2Ω ribbon may be 0.007uV and 30nA. It comes to the same power.

 

[MatthewF]

Thanks PRR, I'm starting to see the whole picture in a much clearer way now, especially the stuff regarding nominal levels.

I realised I must have had something wrong when I connected the output of a -10dBV soundcard (outputting a full scale sine wave) to a scope and saw a ~2v sine wave. Clearly very different to my calculated 0.3v!

>"Just 93dB". Oh you kids.

Yep, fair enough! I was getting carried away with attempting to preserve every microvolt of the signal while forgetting, as you pointed out, that some of it is noise anyway!

Your comments on the dynamic range of home hifi systems and pop music in general help to put this whole issue into context. Suffice to say that I'm now more than happy to use an INA134 as a line level input (and I'm a bit more knowledgeable than I used to be too! :grin: )

Cheers,

Matthew

 

[TedF]

Many years ago, in the days of Ohio Scientific computers, Steve Dove wrote a little Basic routine to calculate the noise power in a resistor across the audio band at NTP (normal temperature and pressure)
It was a little piece of teaching that I never forgot.... it always gave the same result no matter what the value of resistor!
It's a good lesson.... 'Intuition' always seems to lead to clever ways of defeating it, but physics is physics.

Apologies if you thought I was being sloppy..... call it poetic licence to make a point!