Here is a link to the RIAA noise calculator I mentioned a few months ago that I was working on. Its linked to in the 'broadcast blues' thread as well.
The spread sheet is derived from Stuart Yaniger's one with additional inputs to allow the real world noise for up to 5 opamps or discrete amplifiers to be compared.
So Just How Quiet is Your Phono Stage?
!ONLY ENTER DATA IN THE GREEN CELLS!
The spread sheet is derived from Stuart Yaniger's one with additional inputs to allow the real world noise for up to 5 opamps or discrete amplifiers to be compared.
So Just How Quiet is Your Phono Stage?
!ONLY ENTER DATA IN THE GREEN CELLS!
Last edited:
March 19, 2019?
The table heading "Signal to Noise Ratio with input SHORTED ref 1V and ref 0.25V" is in fact not accurate. You have no information for the RIAA network to go on. (Not even Rg, which would make for a decent approximation.) Besides, not sure what the point of this table is even supposed to be. Why 1 V and 0.25V? Why not 2.5 mV and 5 mV?
Formulas in cells V9 and Y9 are referring to J31 and P31, respectively - other cells in these columns are referring to J32 and P32 instead.
'Re(Z) (ohm)' (I31) is actually not 'Cartridge only impedance'.
'|Z| (ohm)' (I32) is not equal to 'Cart and load resistor paralleled impedance' (AE32).
Ultimately, what you want at the end of the day (IMHO) is the RMS sum of:
* opamp voltage noise
* thermal noise of Re{ Z_cart || Rin || -jX_Cin }
* i_n * Re{ Z_cart || Rin || -jX_Cin }
I've been thinking about how one might accommodate amplifiers with extra common-mode current noise from bias current cancellation. As mentioned in the 'broadcast RIAA blues" thread, these have substantially more input current noise when impedances are highly mismatched, as they happen to be in a typical phono MM application. For example, LT1115 i_n is given as 0.9 pA/√(Hz) with matched impedances, but when confronted with 10k vs. 0, this effectively rises to about 3 pA/√(Hz).
At present, my opamp noise calculator does not care one iota about any correlation between input noise currents. 10k + 0 gives the same results as 5k + 5k. Actually, they are literally summed internally, simplifying the calculations.
In a few hours of research of thinking, I've come to the conclusion that
1. one current noise source between inputs is functionally equivalent to two of identical value from each input to ground (this is because you can basically split every ideal current source and put the midpoint at arbitrary potential, and no-one will be the wiser; now it's no longer quite so equivalent if we are looking at the topic of correlation - but this may prove useful when it comes to splitting between differential mode and common mode contributions)
2. my Ib to i_n conversion contained a double fault - I was using a factor of 2 because I erroneously thought this was needed to accommodate the Ib of both inputs; in fact, it is the input current noise of bipolar transistors that is the sum of two terms, one the shot noise of Ib and the other being the shot noise of Ic divided by beta - note that internal current sources and mirrors may add to the Ic noise term! Thankfully, if the opamp is full of noiseless "stuff", this reduces to the shot noise of Ib + Ic/beta = 2 * Ib. So still good for a lower boundary. Phew.
If (2) is true, I may have to distinguish between BJT and FET inputs. So far, I have not seen any indications of significantly more than input gate current being involved. Well, JFETs still have funny complex terms (being capacitive devices) that can interact with complex source impedances. Which is why it is said that they like inductive source Z.
Looks like I will very much have to look into the noise of current sources and mirrors. Imagine your tail current source contributes a certain degree of noise of its own - it will be mirrored with the current across the input pair and appear as correlated Ic noise, on top of the junction-related shot noise already present. That might explain why low-noise designs often rely on plain ol' resistors only (pure conductors have no shot noise).
The table heading "Signal to Noise Ratio with input SHORTED ref 1V and ref 0.25V" is in fact not accurate. You have no information for the RIAA network to go on. (Not even Rg, which would make for a decent approximation.) Besides, not sure what the point of this table is even supposed to be. Why 1 V and 0.25V? Why not 2.5 mV and 5 mV?
Formulas in cells V9 and Y9 are referring to J31 and P31, respectively - other cells in these columns are referring to J32 and P32 instead.
'Re(Z) (ohm)' (I31) is actually not 'Cartridge only impedance'.
'|Z| (ohm)' (I32) is not equal to 'Cart and load resistor paralleled impedance' (AE32).
Ultimately, what you want at the end of the day (IMHO) is the RMS sum of:
* opamp voltage noise
* thermal noise of Re{ Z_cart || Rin || -jX_Cin }
* i_n * Re{ Z_cart || Rin || -jX_Cin }
I've been thinking about how one might accommodate amplifiers with extra common-mode current noise from bias current cancellation. As mentioned in the 'broadcast RIAA blues" thread, these have substantially more input current noise when impedances are highly mismatched, as they happen to be in a typical phono MM application. For example, LT1115 i_n is given as 0.9 pA/√(Hz) with matched impedances, but when confronted with 10k vs. 0, this effectively rises to about 3 pA/√(Hz).
At present, my opamp noise calculator does not care one iota about any correlation between input noise currents. 10k + 0 gives the same results as 5k + 5k. Actually, they are literally summed internally, simplifying the calculations.
In a few hours of research of thinking, I've come to the conclusion that
1. one current noise source between inputs is functionally equivalent to two of identical value from each input to ground (this is because you can basically split every ideal current source and put the midpoint at arbitrary potential, and no-one will be the wiser; now it's no longer quite so equivalent if we are looking at the topic of correlation - but this may prove useful when it comes to splitting between differential mode and common mode contributions)
2. my Ib to i_n conversion contained a double fault - I was using a factor of 2 because I erroneously thought this was needed to accommodate the Ib of both inputs; in fact, it is the input current noise of bipolar transistors that is the sum of two terms, one the shot noise of Ib and the other being the shot noise of Ic divided by beta - note that internal current sources and mirrors may add to the Ic noise term! Thankfully, if the opamp is full of noiseless "stuff", this reduces to the shot noise of Ib + Ic/beta = 2 * Ib. So still good for a lower boundary. Phew.
If (2) is true, I may have to distinguish between BJT and FET inputs. So far, I have not seen any indications of significantly more than input gate current being involved. Well, JFETs still have funny complex terms (being capacitive devices) that can interact with complex source impedances. Which is why it is said that they like inductive source Z.
Looks like I will very much have to look into the noise of current sources and mirrors. Imagine your tail current source contributes a certain degree of noise of its own - it will be mirrored with the current across the input pair and appear as correlated Ic noise, on top of the junction-related shot noise already present. That might explain why low-noise designs often rely on plain ol' resistors only (pure conductors have no shot noise).
Last edited:
The cells that calculate the S/N ratio at 0.25 and 1V are there to give an indication of the noise levels at typical preamp outputs - 0.25V for legacy preamps and 1V for 1dBV. This basically emulates what you would get feeding the preamp into a sound card. The signal to noise ratio given at the top right hand box (by the red box) is calculated with reference to the specified cartridge input voltage. This is what I would use for everyday comparisons.
You don’t need to know the feedback network impedance to get an accurate handle on the total input refereed noise. The cartridge noise and the lower feedback resistor noise - which is a fraction of the feedback network impedance- dominates the noise even at 20 kHz (RMS sum). But, I take your point and will add a cell for putting Rg in - this is typically 50 to 1k (which is on the high side - normally 100 to 220 ohms for noise and network value convenience IMV).
Re mis matched current noise opamps - it’s best just to look to better devices - but it would make an interesting exercise. You will note from the spread sheet that the NE5534A remains remarkably good despite it being 40 years old!
I will take a look at your other points a bit later today - thanks for the feedback.
You don’t need to know the feedback network impedance to get an accurate handle on the total input refereed noise. The cartridge noise and the lower feedback resistor noise - which is a fraction of the feedback network impedance- dominates the noise even at 20 kHz (RMS sum). But, I take your point and will add a cell for putting Rg in - this is typically 50 to 1k (which is on the high side - normally 100 to 220 ohms for noise and network value convenience IMV).
Re mis matched current noise opamps - it’s best just to look to better devices - but it would make an interesting exercise. You will note from the spread sheet that the NE5534A remains remarkably good despite it being 40 years old!
I will take a look at your other points a bit later today - thanks for the feedback.
Basically a good idea, but AFAICT the calculation is normalized to a gain of unity at 1 kHz, vs. the typical 34-40 dB of MM RIAA stages (another piece of information you do not have but probably should). Effectively you are dividing 1 V or 0.25 V by effective input noise with cartridge attached, which (a) makes precious little sense and (b) has nothing to do with "input shorted" at all.
Mind you, I had to fix some nonsense in my opamp noise calculator today, too. Yesterday I boldly wrote
...but in fact that is not true. Input Ib of both transistors is in general uncorrelated, so you also need two uncorrelated noise sources. If you had just one shared between inputs, that would clearly be correlated.
The latter was what my calculator was doing. So it would estimate the noise of unbalanced impedance correctly but be 3 dB too conservative in the case of matched impedances (since i_n * R + 0 = sqrt(2) * RMS(i_n * R/2, i_n * R/2)). I was playing around in LTspice today, which confirmed this. Been like that for years, nobody ever bothered to complain.
Still not actually sure how to best obtain two orthogonal terms to accommodate this correlated noise source business. Can I just add a i_n,cor * |Delta R| term? But even if so, how would one obtain i_n,cor from datasheet graphs? I would have to "un-RMS" the two i_n terms adding up in the unbalanced case, subtracting the uncorrelated noise term (something for which I fortunately wrote a calculator recently). This is because the cases "Rs + Rs matched" and "Rs + 0" are not actually orthogonal - you'd need "Rs + (-Rs)" instead. Now negative impedance is a mess so the two existing cases are going to be as orthogonal as it gets.
I would not sweat about noise arising in opamps that use input bias current cancelling - very few if any of the bipolar input opamps best the 5534A on noise - LM4562, AD797, LME49710, LT1115 all fall down on MM because of their noise current.
If you want a bit lower noise, my recommendation for MM is to go for the OPA1641 (€3). The AD745 is also very good (but expensive - about €10 each last time I looked). For the ultimate in noise, you have to go for LSK389 or better still BF862 if you can find any.
Re the gain point - you don’t need to know gain to calculate noise - that’s the whole idea of using input referred noise voltage. Once you have that and the signal input level, you can calculate the S/N directly.
If you want a bit lower noise, my recommendation for MM is to go for the OPA1641 (€3). The AD745 is also very good (but expensive - about €10 each last time I looked). For the ultimate in noise, you have to go for LSK389 or better still BF862 if you can find any.
Re the gain point - you don’t need to know gain to calculate noise - that’s the whole idea of using input referred noise voltage. Once you have that and the signal input level, you can calculate the S/N directly.
Last edited:
I've made a few updates to the spread sheet and added graphs per Mark's suggestion. (2nd work sheet). Feel free to download it and test drive it. You can enter data for the devices you want to compare in the top left hand area - NB - only enter data in the GREEN cells.
The cartridge and signal level details are entered on the left hand side beneath the main table.
The cartridge and signal level details are entered on the left hand side beneath the main table.
That is really a cool spread sheet you have there. I did notice one thing that is strange.
The graph showing the "S/N ratio vs signal level" shows the LSK389 as being the worse of the lot.
However in the graph above it shows LSK389 to be slightly worse than the BF862, situated between the BF862 & the opamps.
I assume that if one puts two jfets in parallel that the "env" would be reduced by sq rt of 2? So for the lsk389 it would be 1.27/1.414 = 0.9 which is the same as the BF862?
The graph showing the "S/N ratio vs signal level" shows the LSK389 as being the worse of the lot.
However in the graph above it shows LSK389 to be slightly worse than the BF862, situated between the BF862 & the opamps.
I assume that if one puts two jfets in parallel that the "env" would be reduced by sq rt of 2? So for the lsk389 it would be 1.27/1.414 = 0.9 which is the same as the BF862?
"What quickly becomes apparent is that even with a phono amplifier with say 90 or 100 dB S/N, you will not get better than 71 or 72 dB S/N with the typical phono cartridge ref 3mV input."
That is why the old IHF standard A-202 (now CEA-490) required that preamp noise be measured with a standard cartridge terminated with a standard load. The standard cartridge was 500 mH in series with 1 kOhm, (Shure M-75) and the standard load was 47 kOhms paralleled by 125 pF. A noise calculation of the standard cartridge and load gives an A weighted noise of about 430 nV over the 20-20kHz bandwidth. That is about 78 dB below the standard reference level of 5mV at 1 kHz. That is the best you can do with a noiseless preamp. Why the IHF standard required A weighting I don’t know, but it makes the SNR look better than if you just sum the noise without weighting. Without weighting the noise sums to about 600 nV. If you do a Spice noise analysis on a phonograph cartridge you will see that about half the noise is in the first decade, 20-200 Hz and the A weighting greatly attenuates this noise. The A weighting is almost 20 dB down at 100 Hz.
That is why the old IHF standard A-202 (now CEA-490) required that preamp noise be measured with a standard cartridge terminated with a standard load. The standard cartridge was 500 mH in series with 1 kOhm, (Shure M-75) and the standard load was 47 kOhms paralleled by 125 pF. A noise calculation of the standard cartridge and load gives an A weighted noise of about 430 nV over the 20-20kHz bandwidth. That is about 78 dB below the standard reference level of 5mV at 1 kHz. That is the best you can do with a noiseless preamp. Why the IHF standard required A weighting I don’t know, but it makes the SNR look better than if you just sum the noise without weighting. Without weighting the noise sums to about 600 nV. If you do a Spice noise analysis on a phonograph cartridge you will see that about half the noise is in the first decade, 20-200 Hz and the A weighting greatly attenuates this noise. The A weighting is almost 20 dB down at 100 Hz.
Good points. I was unaware of the IHF test standard re phono carts.
John Atkinson (Stereophile editor) was on here a few years ago and one of the members pointed out that measuring the noise levels by simply shorting the phono amp inputs together was as good as useless.
I have a AT95 that I bought for about $35 that I won’t use for listening to music, so I will press it into service as a noise source.
John Atkinson (Stereophile editor) was on here a few years ago and one of the members pointed out that measuring the noise levels by simply shorting the phono amp inputs together was as good as useless.
I have a AT95 that I bought for about $35 that I won’t use for listening to music, so I will press it into service as a noise source.
Last edited:
Thanks for the details of A-202.
> Why the IHF standard required A weighting I don’t know,
Because they assumed that "noise" would be played soft, where the ear does not hear bass or high-highs, and A-weight is a fair and convenient approximation of ear response at low levels. Your "half in 20-200Hz" is real but not audible and should be discounted.
IMHO, using "A" to weight the damage or annoyance of LOUD sounds, as is often done today, is mis-use. "A" was invented for soft sounds.
> Why the IHF standard required A weighting I don’t know,
Because they assumed that "noise" would be played soft, where the ear does not hear bass or high-highs, and A-weight is a fair and convenient approximation of ear response at low levels. Your "half in 20-200Hz" is real but not audible and should be discounted.
IMHO, using "A" to weight the damage or annoyance of LOUD sounds, as is often done today, is mis-use. "A" was invented for soft sounds.
Yes, my understanding is that it was designed to show that our loudness perception of sound level drops off at frequency extremes. Agree it is an abused measure for industrial/traffic noise and its used to allow companies and individuals to get away with noise pollution. Rod Elliot has a piece on this.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.