Measuring noise and SINAD of MM phono preamps properly

[Heinrich]

This is the updated chart with SINAD and noise results ...

Using the table I calculated the self-noise of your Shure M35X cartridge. It is around -80dB ref to 5mv (taken from the OPA627 results, no In assumed). Its contribution is quite significant at just a little bit below the amp's voltage noise. As always, the load 47kOhm resistor is accounted to the cartridge ;-)

Besides that debate, thank you so much for substantially clearifying the basic issue. Now we only have to spread the word, as they say.

The noise figures:
OPA627: 3.1dB
NE5534: 4.7dB
Difference is 1.6dB which in comparison to expected tolerances in device properties and further influences is neglegible. The price difference is not as the OPA627 was about 60 times as expensive as the NE5534. As far as I know only fake OPA627 is available today, out of production for long.

 

[narshorn]

This is the updated chart with SINAD and noise results for the Openamp MM phono preamplifier with various op-amps. Noise was measured with M35X cartridge at the input. SINAD was measured from 50 ohm low noise generator and 5mV/1kHz signal. SINAD with the cartridge was calculated using the formula posted in the post #1 of this thread.
4 op-amps with bipolar input and 2 opamps with JFET input were tested. Measured results for LT1028 are not shown, because I was not able to make it stable in the phono circuit yet.

View attachment 428007

When measured with a 50 ohm generator (which is almost same method as used at ASR with 20 ohm AP generator), the best SINAD result was for ADA4898, 85dB. But, when the cartridge was connected to the input, this op-amp had the highest noise and thus the worst SINAD in real conditions. This is for the reason that the generator method does not take into account the input current noise of the preamp. For this reason, the method with low impedance generator is only valid for low impedance signal source, but not for the high impedance source as is the MM cartridge loaded by 47k standardized impedance. This is what I have been trying to explain, with higher or lesser success, last weeks. JFET op-amps with with reasonably low voltage noise (<8uV/rt(Hz)) are much better choice for MM preamps as they are free of input current noise issue. Kudos to the old NE5534! (bipolar). As Douglas Self states, this is the preferred one from bipolar opamps.

Please also note excellent conformance of simulation and measurements for JFET op-amps. Not so perfect for bipolar op-amps, due to a bi unpredictable value of the input current noise of individual parts.

--------------

Oscillations with LT1028:
View attachment 428016

This results in highly elevated noise floor and SINAD (with generator) falls down to 67 dB.

Hi pma,

Thanks you for all those measurements
Yes, NE5534 stays top cream for bipolars

So, no data of LT1028 with M35X loading ? Maybe better than with the generator load ...

To cure the oscillation with those sensitive high GBW devices, maybe adding an extra small decoupling capacitor directly between V+ and V- pin of the device
- a ceramic good quality one, in the range 47-100 nF, and with the shortest leads possible (just under, or just on top of the device) - would help

Best regards,

n.

 

[Heinrich]

Using the table I calculated the self-noise of your Shure M35X cartridge. It is around -80dB ref to 5mv (taken from the OPA627 results, In=0 assumed).

Btw: all these measurements and calculations would look different when sporting a moving iron (MI) or a high output moving coil pickup. The coil's inductance is decidely lower with such devices.

 

[lini]

Btw: all these measurements and calculations would look different when sporting a moving iron (MI) or a high output moving coil pickup. (...)

Well, the latter for sure - but regarding MIs your statement would appear too generalised, as there are many MI models with rather high inductance/impedance.

Greetings from Munich!

Manfred / lini

 

[pma]

Shure M35X "cartridge noise" measured with various op-amps. The noise is referred to the amp input, thus cartridge output. It is still the same song, only different method of measurement and display. NE5534 not used as it sits now in another preamp. It would just a bit worse than OPA627 but better than AD797. LT1028 did not meet expectations. Yes no oscillations.

____________________________

Total noise at cartridge output, simulated vs. measured:

Op-amp	"cartridge noise" measured	"cartridge noise" simulated
AD797	7.053 uV	10.690 uV
OPA627	2.343 uV	3.151 uV

 

[Heinrich]

Shure M35X "cartridge noise" measured with various op-amps.

Op-amp	"cartridge noise" measured	"cartridge noise" simulated
AD797	7.053 uV	10.690 uV
OPA627	2.343 uV	3.151 uV

We are looking for an easy to understand indication of signal degradation, right? The SINAD could be the output noise + harmonic distortion compared to the standard 1kHz full scale (=max. volume) at the output.

With the 'cartridge noise' shown here I recalculated numbers.

S/N = 20 * log10( 2.343µV / 5 mV ) = -66,6dB

The result doesnt match expectations, neither measured output as given a few posts above. I might speculate, the RIAA attenuation is not taken into account, which is o/k from certain points of view. Aren't we complicating things a little with this approach?

In post #81 I introduced the 'noise figure'. The aim is to compare the actual outcome after amplification with a real device to what would come out if the amp was ideal, no noise or distortion. As such it is output related, the RIAA equalization would be included.

Even more so, the noise figure would be based on the notion that the signal generator is also an imperfect device. We want to isolate the amp's contribution to the total signal degradation, noise current included, from the generator's imperfections.

In our case the assembly of pickup resistiance, inductance, load resistor and shunt capacitance poses the signal generator. (Of, course, 'electronic cooling' of load is possible, but not really practical.) The signal generator, if taken as such, generates thermal noise on its own, that is going to be amplified. When connected to the amp, the latter would add noise voltage, and if there was noise current at the input, it is converted at the output impedance of the signal generator to voltage noise, that is amplified also.

I recalculated the numbers given in post #78. The input noise current of the OPA627 is assumed to be neglegible in first approximation. Then we are left with just two noise sources, the amp's voltage noise, and the voltage self-noise of the pickup assembly, load resistor included. The 50Ohm generator gives us the amp's voltage noise alone, the measurement with a Shure M35X generator gives us noise as a combination like N = sqrt( Namp^2 + Ngen^2 ). From that it is easy to calculate Ngen, in this case -80dB. It matches the difference 50Ohm to Shure M35X to be 3dB; generator noise and amp noise contribute in the same proportions.

Not the least, the noise figure motivates to consider the self-noise of the generator as a limit. It would tell what the maximum of SINAD could be with an ideal amp, because the signal is already degraded by the intended generator, the pickup assembly namely, at least in terms of noise. With standard pickups that would be -80dB give or take. And you proved it.

As always, thanks for your kind contribution!

 

[pma]

Just for your information, I would like to repeat that the noise measured with the generator connected compared to situation when the input of the preamp is terminated by 50 ohm is the same. Please take it as a fact. The reason is that the “generator” is DAC with output noise of 2 uV followed by voltage divider 1 : 55, with 50 ohm output impedance. This reduces “generator noise” to 36 nV (2uV divided 55x), or 0.25nV/rt(Hz) noise density if you wish. Output voltage of the generator for 5 mV preamp input is 5mV x 55 = 0.275V. 0.275V to 36nV gives generator signal SNR of 137.66 dB. This is definitely good enough. Much more is the thermal noise of the 50 ohm resistor, regardless if generator is connected via the divider. 50 ohm resistor thermal noise voltage over 20kHz band is 130nV, 0.13uV. Again, please do not overlook the generator output divider, which reduces generator noise to 36nV. The “generator” does not add noise here, the contribution is unimportant, increases 130nV thermal noise of 50 ohm resistor to 135nV. Also, please refer to post #85 and you will see that I am able to measure reliably voltage noise density as low as 4nV/rt(Hz) and lower. 4nV/rt(Hz) is just the thermal noise density of Shure M35X cartridge at low frequencies (970 ohm resistance).

 

[Skeeter]

This is the updated chart with SINAD and noise results for the Openamp MM phono preamplifier with various op-amps. Noise was measured with M35X cartridge at the input. SINAD was measured from 50 ohm low noise generator and 5mV/1kHz signal. SINAD with the cartridge was calculated using the formula posted in the post #1 of this thread.
4 op-amps with bipolar input and 2 opamps with JFET input were tested. Measured results for LT1028 are not shown, because I was not able to make it stable in the phono circuit yet.

View attachment 428007

When measured with a 50 ohm generator (which is almost same method as used at ASR with 20 ohm AP generator), the best SINAD result was for ADA4898, 85dB. But, when the cartridge was connected to the input, this op-amp had the highest noise and thus the worst SINAD in real conditions. This is for the reason that the generator method does not take into account the input current noise of the preamp. For this reason, the method with low impedance generator is only valid for low impedance signal source, but not for the high impedance source as is the MM cartridge loaded by 47k standardized impedance. This is what I have been trying to explain, with higher or lesser success, last weeks. JFET op-amps with with reasonably low voltage noise (<8uV/rt(Hz)) are much better choice for MM preamps as they are free of input current noise issue. Kudos to the old NE5534! (bipolar). As Douglas Self states, this is the preferred one from bipolar opamps.

Please also note excellent conformance of simulation and measurements for JFET op-amps. Not so perfect for bipolar op-amps, due to a bi unpredictable value of the input current noise of individual parts.

--------------

Oscillations with LT1028:
View attachment 428016

This results in highly elevated noise floor and SINAD (with generator) falls down to 67 dB.

Excellent work PMA - his is exactly what I was referring to in my earlier discussion on this subject.

 

[Skeeter]

 

[Heinrich]

Just for your information, I would like to repeat that ...

I'm afraid I cannot understand the importance of your explanation. But you stated already, that you won't consider the 'noise figure' as a worthwhile measurement, and that the self-noise of the pickup is of no further interest. It's your thread. In case you would like to switch opinions and discuss before publishing, send me a personal message. I unwatch the thread.

 

[pma]

So please show me where is the noise added from the generator, if we compare measurements with 30dB gain amp with input terminated by 50 ohm terminator and the same amp driven from the generator that is used in this thread. Measured noise is the same, within the statistical error. Where is the generator noise contribution?

 

[Heinrich]

So please show me where is the noise added from the generator, if we compare measurements with 30dB gain amp with input terminated by 50 ohm terminator and the same amp driven from the generator that is used in this thread. Measured noise is the same, within the statistical error. Where is the generator noise contribution?

View attachment 428272 View attachment 428273

I think my post #86 elaborates on this extensively, as a final word on what was discussed before, from my side, though. Douglas Self, as an authority, not that I care to much, starts his discussion of MM preamps with introducing the 'noise figure', hence 'self-noise' (no pun); it was common practice when I was way younger and on topic more than today. Can't do more. That doesn't overcast your effort, no worries--thank you! To the contrary, I would like the current noise aspect to be acknowledged. We shall not spoil won't blurr this perspective with another, still valid, open controversy.

 

[NTTY]

First, we are not looking for accuracy in low frequencies in this discussion. Second, 32K points gives you near 1 Hz resolution already.

But these are not the main issue. The issue is that you seem to be saying, "look at how low the noise floor is" but such a statement cannot be made without looking at impact of FFT gain. Your 128K FFT divides the noise buckets by four, resulting in low numbers on the graph when in reality that doesn't happen.

Hi Amir, visually yes, but in the computation of THD+Noise, no. Too small FFT length AND only few averages, will show variations in noise calculation and therefore THD+N, between measurements. So increasing the FFT shall not decrease the noise floor when computing the SINAD, unless the initial FTT size was too low, ie making the computation less precise. Same goes with increase of averages, by the way, but less so.

Couple of examples to illustrate. Like @restorer-john, I used an old Denon preamplifier (PRA-S10 in my case) with its MM phono input. All measurements below were done at 48kHz sampling rate input of the ADC, output being 50mV and input 2V (gain 32.05dB).

Let's start with 2M FFT - 32 averages:

SINAD = 82.6dB

Same but 256k FFT - 32 averages:

SINAD = 82.6dB

Same but 128k - 32 averages:

SINAD = 82.6dB

Same but 64k FFT - 32 averages:

SINAD = 82.6dB

Every time we've seen the noise floor increasing visually, but the calculated noise, and the resulting THD+N, remained the same at -82.6dB.

Now same again with 32k FFT - 32 averages:

SINAD = 81.1dB (best score when measuring at this FFT size).

Only here, the calculated noise floor is increased, and that's because of the lower resolution, not because there's more noise.

As a matter of facts, reducing the averages to 4 instead of 32, but keeping the 32k FFT size, I see variations from this:

SINAD = 79.4dB

to that:

SINAD = 82.7dB

So the combination of (too) small size FFT and low number of averages shows significant variations in the results because of the lower resolution. So the SINAD will vary significantly depending on the moment of the capture. The operator can perform multiple captures and keep the best one, but alternative option is to increase FFT size and averages.

Indeed, with a 1M FFT size, and only 4 averages, the results are more precise with much less variation between two screen captures:

Example 1, 1M FFT size - 4 averages:

Example 2, 1M FFT size - 4 averages:

It is only 0.5dB difference between best and worst capture, as opposed to 3.3dB with 32k FFT.

All of that is the reason why, as per my own rule, I always go for an FFT size which is the next higher length than 2 times the sampling rate (128k FFT for 48kHz sampling rate, for instance), and 32 averages. It only takes a few seconds to compute and I don't bother waiting for the best or worst measurement.

Cheers

 

[Heinrich]

The Burr-Brown JFE2140 with 0.9nV/√Hz and 1.6 fA/√Hz is much better for MM preamps than any of the opamps you listed

Thank you, astounding as long as one thinks it is an opamp ;-) I will consider it for my guitar as a piezo frontend.

But anyway, it shows what can go wrong when neglecting the limits dictated by the self-noise of the pickup. As exemplified above a regular MM would exhibit noise of around -80dB anyway. The -88dB as depicted in the first image of your attachments cannot be achieved in principle. Total noise may be -79dB, but good old NE5534 would yield only 3dB worse with less effort by a big margin. Never mind,

 

[restorer-john]

output being 50mV and input 2V (gain 32.05dB).

5mV.

 

[Heinrich]

I use "Passive Cooling" in my preamp ...

Hi Nick, as before I can only tell that we possibly won't distract from the importance of noise current. There were some devices that fell into the trap of evaluating noise with very low source impedances alone.

Your 'passive cooling' is something else, and yes, it supports my case of self-noise, but that's a different one. As I'm , alas, not proficiant in your language, I connot make out the technology. Shifting the upper pole by using very small capacitane, raising the Q by a large resistor, then bring its effect down by an appropriately tailored Linkwitz biquad (we know that for high passes , switch to low pass?). Thats worth a dedicated thread, indeed.

 

[undeakyetonde]

The Burr-Brown JFE2140 with 0.9nV/√Hz and 1.6 fA/√Hz is much better for MM preamps than any of the opamps you listed
[

]

Mr @Michael Fidler do you mind sharing your thoughts on this? Are there any drawbacks to building input stage of a MM amplifier based on JFE2140? I reckon you considered a lot of different options

 

[Michael Fidler]

Mr @Michael Fidler do you mind sharing your thoughts on this? Are there any drawbacks to building input stage of a MM amplifier based on JFE2140? I reckon you considered a lot of different options

It's a perfectly valid way of making quick measurements if you have a soundcard. I used to use Audacity for this. The one thing missing is measuring the noise with a cartridge connected, as connecting the phono preamp to a low-impedance line output only tells half the story.

JFE2140 has very low noise, but is also a FET device and hence will exhibit non-linear capacitance when faced with even a relatively small amount of common-mode voltage. It is therefore probably a good idea to cascode it with a pair of BJTs. You also have to consider RFI rejection which is typically much worse for discrete circuits.

 

[Heinrich]

It's a perfectly valid way of making quick measurements if you have a soundcard. I used to use Audacity for this. The one thing missing is measuring the noise with a cartridge connected, as connecting the phono preamp to a low-impedance line output only tells half the story.

Right and not so. True is the very fact, that measuring with a 50Ohm generator doesn't represent the common use case, and it suggests, falsely, a way to good s/n-ration. Not that simple is the evaluation of the cartridge as the signal generator.

With FET we have neglegible influence of the noise current. Hence all of additional noise when connecting a cartridge is from the thermal self-noise of the generator itself - it is a resistor with complications, but a resistor, the mandated 47kOhm load. And that should be considered as a given and a constant for all preamps in comparison. It limits even the theoretical optimum.

Sidenote: the capacitance induced distortion of a FET is always utterly neglegible in relation to the distortion of the medium from recording, mixing, via cutting, pressing up to the final pickup process. Common mode voltage is about 0.005V here, not 4V as in (analog) mixing consoles. It goes with the square in voltage swing, right?

 

[Michael Fidler]

Right and not so. True is the very fact, that measuring with a 50Ohm generator doesn't represent the common use case, and it suggests, falsely, a way to good s/n-ration. Not that simple is the evaluation of the cartridge as the signal generator.

With FET we have neglegible influence of the noise current. Hence all of additional noise when connecting a cartridge is from the thermal self-noise of the generator itself - it is a resistor with complications, but a resistor, the mandated 47kOhm load. And that should be considered as a given and a constant for all preamps in comparison. It limits even the theoretical optimum.

Sidenote: the capacitance induced distortion of a FET is always utterly neglegible in relation to the distortion of the medium from recording, mixing, via cutting, pressing up to the final pickup process. Common mode voltage is about 0.005V here, not 4V as in (analog) mixing consoles. It goes with the square in voltage swing, right?

Yes, nonlinear junction capacitance is proportional to the square of voltage. But there are two other mechanisms at play here:

1. The nominal output increases rapidly with frequency, the cartridge being a velocity-to-voltage converter, so we can expect at least 10 times more voltage than 0.5mV for a given displacement at 10kHz vs the quoted 1kHz. Surface clicks and pops also produce high transients. If we want 20dB of overload margin, our preamp must be capable of handling at least 250mV at 10kHz.
2. The output impedance of an MM system rises rapidly as cartridge inductance dominates, contributing up to 31 kilo-ohm at 10kHz for 500mH, further exacerbating the non-linearity. Well over 100 times greater than a typical generator output.

If the source was only 5mV at a few hundred ohms, this wouldn't be a problem, but with 100mV or more at 31k it's an entirely different story.