Confused question is confused.
What do you call "line level" (400 mV? 2 V? More?), and what's your "application"?
When you want a 22k input impedance and settle on a 22k/22k inverting amplifier to achieve it, you'll hard a hard time beating a TL072 by more than 1-2 dB. You probably don't want one in the 16.5 dB line amplifier stage of your preamp though, or a number of other places. 5 nV/√(Hz) isn't a bad rule of thumb, it's the equivalent of 1.5 kOhms and as such low enough for typical surrounding impedances in a lot of typical audio circuits that you might encounter. It is also typical for the first wave of "low noise" audio opamps (NE5532, LM833, MC33078). There is nothing inherently important or magical about this number.
In order to be universally of use, input noise must be backed up by output stage driving abilities. Many older very-low-noise opamps assumed a high-gain circuit like a tape head or microphone preamplifier and wouldn't have the oomph to back up a low-gain noninverting amplifier at line level that actually gets close to their rated noise. The likes of NJM2043 or NJM2122 come to mind. Thus a general purpose part rarely needs to be ultra-low-noise; that kind of dynamic range generally just isn't required, and corresponding power levels may be undesirably high (resistor loading, power consumption). Dynamic range is, after all, the ratio of signal power to noise power.
What do you call "line level" (400 mV? 2 V? More?), and what's your "application"?
When you want a 22k input impedance and settle on a 22k/22k inverting amplifier to achieve it, you'll hard a hard time beating a TL072 by more than 1-2 dB. You probably don't want one in the 16.5 dB line amplifier stage of your preamp though, or a number of other places. 5 nV/√(Hz) isn't a bad rule of thumb, it's the equivalent of 1.5 kOhms and as such low enough for typical surrounding impedances in a lot of typical audio circuits that you might encounter. It is also typical for the first wave of "low noise" audio opamps (NE5532, LM833, MC33078). There is nothing inherently important or magical about this number.
In order to be universally of use, input noise must be backed up by output stage driving abilities. Many older very-low-noise opamps assumed a high-gain circuit like a tape head or microphone preamplifier and wouldn't have the oomph to back up a low-gain noninverting amplifier at line level that actually gets close to their rated noise. The likes of NJM2043 or NJM2122 come to mind. Thus a general purpose part rarely needs to be ultra-low-noise; that kind of dynamic range generally just isn't required, and corresponding power levels may be undesirably high (resistor loading, power consumption). Dynamic range is, after all, the ratio of signal power to noise power.
It really depends on what you're ultimately doing! If I'm designing for low noise it's either for high gain (non audio, more measuring small signals) or pure specmanship. In which case, yeah, 15 nV/rtHz input referred is way worse than many opamps will do I'm circuit.
You have to ask yourself how your gain structure looks and what effect you'll ultimately have on end-to-end SNR. Audibility is pretty easy: nothing playing and a reasonably quiet room: do you hear anything? No? You're good.
You have to ask yourself how your gain structure looks and what effect you'll ultimately have on end-to-end SNR. Audibility is pretty easy: nothing playing and a reasonably quiet room: do you hear anything? No? You're good.
That graph looks like a graph from an article about resistor 1/f noise measurements (which some people call current noise because it only occurs when there is current flowing through the resistor, which can easily cause confusion with an amplifier's equivalent input noise current).
In any case, when you want to accurately measure the noise of 10 kohm resistors, both 1/f and thermal noise, it is convenient when you have a measuring amplifier with lower noise. The thermal noise of 10 kohm at room temperature is about 13 nV/sqrt(Hz), so 5 nV/sqrt(Hz) is comfortably below that.
Thanks. With line level I meant a situation where you hardly need more than 12dB. But for some situations you might need higher value resistors that can become the weakest link in the circuit.
If you use AD797 in place of the above TL072 buffer, you don't want such a high impedance, and most sources are capable of handling such situation.
My question is trying to put noise numbers into perspective. People discussed about low noise resistors and somebody chimes in "Hey, I use noisy carbon resistors and I'm fine!". This thread is talking about low noise transistors, but do you think people can ABX low noise transistor and regular transistor?
For amplifier input stage, I have used all kinds of low noise transistors so I have no question. But opamps and line level applications are rather something new for me.
Yes, the image is from the Ligo:
https://dcc.ligo.org/public/0002/T0900200/001/current_noise.pdf
I'm trying to figure out the minimum noise threshold, especially for opamps, so I can choose the right opamp for certain application. If 5nV/rtHz (e.g. at 10kHz) is the threshold then I will be stucked with few options. OPA627 was something I own and conformed with that requirement but not something I would like to use.
Thanks.
The Signal to noise ratio depend on output impedance of the source signal and value of the resistor feedback of the op-amp.
The best S/N is when output impedance of the source signal = voltage input noise / current input noise.
More small value of resistor feedback more better S/N, but more distortion.
I have those covered. Except what does it mean (to the ears) when the noises differ by a certain amount. If I can set a threshold for the noise, I will have more options to prioritize other parameters than noise.
Interesting. I will use this method to define my noise threshold. Thanks.
It all depends on what you want. Just three examples:
Suppose you want to buffer a 2 V RMS line level signal from a DAC and you want the noise of the buffer to be at least 130 dB(A) below the signal. You can then allow 2 V/10^(130/20 ~= 0.63246 uV of noise within the bandwidth of an A-weighting filter. The noise bandwidth of an A-weighting filter is about 13 kHz (or actually a bit more or a bit less depending on what happens above 20 kHz). The allowable noise density is then 0.63246 uV/sqrt(13000 Hz) ~= 5.547 nV/sqrt(Hz).
If you consider 110 dB(A) to be good enough (which is still about 15 dB lower than the dithered quantization noise of a CD), 55.47 nV/sqrt(Hz) will do.
As another example, suppose you have a main amplifier with 100 W per channel output power, 1 V input sensitivity and you want the noise to be below 0 dB(A) sound pressure level when using 90 dB, 1 W, 1 m loudspeakers in a room with 1 m noise radius. The noise of each channel has to be below one half of this, which means -3.0103 dB(A) SPL. The noise power into each loudspeaker is then -93.0103 dBW(A). That's 113.0103 dB(A) below the maximum power of the amplifier, so its equivalent input noise has to be 113.0103 dB(A) below its sensitivity level of 1 V RMS.
That boils down to 2.2361 uV RMS noise over 13 kHz -> 19.611 nV/sqrt(Hz) equivalent input noise of the main amplifier.
Suppose you want to buffer a 2 V RMS line level signal from a DAC and you want the noise of the buffer to be at least 130 dB(A) below the signal. You can then allow 2 V/10^(130/20 ~= 0.63246 uV of noise within the bandwidth of an A-weighting filter. The noise bandwidth of an A-weighting filter is about 13 kHz (or actually a bit more or a bit less depending on what happens above 20 kHz). The allowable noise density is then 0.63246 uV/sqrt(13000 Hz) ~= 5.547 nV/sqrt(Hz).
If you consider 110 dB(A) to be good enough (which is still about 15 dB lower than the dithered quantization noise of a CD), 55.47 nV/sqrt(Hz) will do.
As another example, suppose you have a main amplifier with 100 W per channel output power, 1 V input sensitivity and you want the noise to be below 0 dB(A) sound pressure level when using 90 dB, 1 W, 1 m loudspeakers in a room with 1 m noise radius. The noise of each channel has to be below one half of this, which means -3.0103 dB(A) SPL. The noise power into each loudspeaker is then -93.0103 dBW(A). That's 113.0103 dB(A) below the maximum power of the amplifier, so its equivalent input noise has to be 113.0103 dB(A) below its sensitivity level of 1 V RMS.
That boils down to 2.2361 uV RMS noise over 13 kHz -> 19.611 nV/sqrt(Hz) equivalent input noise of the main amplifier.
Sometime. But for line pre-amp, distortion is much importance than signal to noise ratio, because input signal level is high. You just need very good voltage regulator.
So may be this is the background of the 'standard' used by Groner? Or he just used NE5532 as reference. The weakest link in my system is not the DAC.
Hmmm great! I was expecting 45nV/rtHz as the worst case.
Nice, as I'm shooting at 15nV/rtHz (10kHz). Thanks!
You're welcome: it's really difficult for any of us to say what is an acceptable noise profile for you, given the rest of your setup, background noise, listening levels, etc. What I suggested is a torture test, since masking/etc from actual playback are going to affect your ability to hear noise, except long silences.
There is a question whether a preamp is ever needed in modern audio system. Most audiophiles think that you don't need it (they prefer passive pre-amp instead). The fact is that a preamp has to be extremely good in order to be considered acceptable.
Yes, I know everyone may have different standard, that's why I'm curious where Groner get his high standard from. Based from experience in the past, I have never had a preamp that is "noise-free". The added noise and the distortion is always very audible. And I want to make one which can be considered "wire with gain"
The power supply will be something similar to the Super Regulator (Usually I used pre-reg with LM317/337 but never used AD797 as the error amp, only OPA627)
Most audiophiles do not know about engineering. May be they are better in listening test (not always, some lies) but their conclusion may be wrong.
You do not need pre-amp if the input level can make your amplifier clipping.
Passive pre-amp is good if you put it very close to your amplifier. Output impedance of the passive pre-amp usually high, depend of the attenuator's position.
Impedance, generally, or rather: its real part.
Be warned that this is a somewhat complex (ahem) subject that can be rather nonintuitive. Condenser microphone biasing is a good example. When your source essentially is a capacitor in the pF range, funny things start happening. While it probably is not too hard to imagine that input capacitance forms a voltage divider much like source and input resistance and as such too much of it reduces signal levels (not good for S/N for obvious reasons), input biasing is another issue. These circuits typically use bias resistors in the 10 gigaohm range (no typo). You could use a 1Meg or so and frequency response would actually still be OK - but noise would be through the roof. There are two ways of explaining this: (a) the 1Meg has way more noise current, which the whole impedance turns into a noise voltage (V = Z * I), or (b) while the 10G resistor has 100 times the unloaded noise voltage, the source resistance of that noise source is 10000 times as high, and suddenly the impedance of a (few) dozen pF isn't actually all that high in comparison, allowing condenser source impedance to dominate - so that noise is actually reduced 100 times. Considerations like these have led Douglas Self to pursue synthetic input impedances, e.g. using a 1Meg but use active circuitry to make it look like a 47k (but with the noise current of a 1Meg!).
Another example is that of predominantly inductive sources, which like predominantly capacitive (e.g. FET) inputs, resulting in part of the noise cancelling out.
Last edited:
Of course it is true. Most of signal source at line level have low impedance output. If the signal level can clip your amplifier, it just need attenuator (passive pre-amp).
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.