You are not alone, I was never able to confirm these measurements and they are also contradicted by the full schematic provided in the Proceedings of the ISSCC. I never measured a correlation coefficient better than 20% and if you examine the statistics claims of near 100% correlation are not physically real.
Last edited:
Your calculation does not differ from mine, both result in 3.0pA/rtHz (one decimal in the final result is usually enough) when using 33nV/rtHz@1Khz.
I have always found that the significant digit stuff children learn at school is complete and utter nonsense, hence my use of more digits - but that is very much of topic. Besides, using lots of digits is handy when you want to know whether two calculations are equivalent - you may get the same two digits by chance, but it is unlikely that you get the same ten digits by chance.
Actually I found 1.377465845 pA/sqrt(Hz). From the context I gather that that 1 pA/sqrt(Hz) refers to the differential noise current. Hence, you can represent it with one current generator between the positive and the negative input and the current gets multiplied by 2 Rs. That gives you sqrt(2) times the noise voltage that two independent noise current sources at the inputs would give you.
Nonetheless, the graph definitely doesn't match the 1 pA/sqrt(Hz) figure.
Last edited:
You would be the ideal contender to find the next prime
The mistake you made, is that you assumed that the 1pA/rtHz base currents are fully correlated on both sides, thereby giving 20nV/rtHz for the two 10k resistors.
However these noise currents are of course 100% uncorrelated and produce only 14nV/rtHz.
This is also wrongly presented in Linear's fact sheet on page 11.
So I think Linear's graphs are produced and not measured by some marketing dept.
Hans
The Ib compensation currents are produced by a split collector pnp and the correlation coefficient has been a source of argument for over 30yr. Emitted carriers are statistically independent so I don't see the argument for 100% correlation holding water.
Last edited:
Scott,
Thanks for your contribution.
This subject seems like a small pot of worms, however a few things can be said with full confidence for amps like the LT1028 with input current compensation.
When uncorrelated 1pA/rtHz current noise is injected from both inputs into two 10 Kohm resistors, an input noise contribution of 14nV/rthz will be added on top of resistor noise and voltage noise.
When fully correlated 1pA/rtHz noise will be injected into two 10 Kohm resistors, each side apart will see 10nV/rtHz because of this.
But being subtracted at the + and - input, the result will be zero because of their correlation.
That is why the LT1028 spec on page 11 is dead wrong, because Linear simply adds both noises to 20nV/rtHz.
When having two 10nV/rtHz sources at both inputs, input noise contribution can be anything between zero and 14nV/rtHz, depending on their degree of correlation, but never more than 14nV/rtHz.
When using unmatched source resistors things are getting complicated, because correlated noise from current compensation circuitry is taking its part.
However, I'm fully with you that it is quite unlikely to be more than a small additional fraction.
Spice doesn't bring us any further either.
The AD797 model, also having input current compensation, uses exactly the same 2.2 pA/rtHz current noise for matched and unmatched sources.
But the LT1028 shows in LTSpice 1.4nV/rtHz for matched 10K sources and 2.9nV/rtHz for a single 10K resistor.
This is yet another set of figures that do not match their specs.
However to be honest, in the low ohmic environment where these amps are normally used, this uncertainty is unlikely to cause problems.
Hans
Last edited:
I think you should make a distinction between the physical, the mathematical and the commercial.
PHYSICAL
If you look at what physically happens inside the op-amp, you will see that there are many random processes going on that contribute to the noise current. Some of these, such as the base current shot noise of the input transistors and the collector current shot noise of the output transistors of the base current compensation circuit, are independent and hence uncorrelated for the positive and the negative input. Other contributions, such as those from most other noise processes inside the base current compensation circuit, end up essentially the same in both inputs and cancel when the inputs are driven from equal impedances. (These latter contributions are quite large because the model transistor in the base current compensation circuit runs at a much lower current than the real input transistors, and an amplifying PNP mirror has to amplify the base current up to the correct level.) So far, so good.
MATHEMATICAL
If you look at it from a higher level of abstraction, you see a thing with a current i1(t) flowing into the positive input and a current i2(t) flowing into the negative input. This can be mathematically represented as two current sources with value icm(t) = (i1(t) + i2(t))/2, one from each input to ground (or supply), plus a differential current idm(t) = (i1(t) - i2(t))/2 flowing from the positive to the negative pin. This is just a matter of adding and subtracting, it is not a physical model of what's happening inside the op-amp.
When you want to relate this to the physical representation, you will find that the noise sources that inject independent noise currents into the positive and negative inputs end up in icm(t) as well as in idm(t). In fact 1/sqrt(2) of their spectral density (half their power spectral density) ends up in the spectral density of icm and 1/sqrt(2) in the spectral density of idm.
COMMERCIAL
The honest thing to do for Linear Technology would be to specify the noise current with very different driving impedances for the positive and negative inputs, because most real-life low-noise op-amp circuits have very different driving impedances for the positive and negative inputs. The fact that they don't put 3 pA/sqrt(Hz) or 3.25 pA/sqrt(Hz) or whatever it may be in the tables shows that they've been looking for tricks to get smaller noise current numbers into their datasheet.
If the aim is to minimize the number without risk of getting sued, it is more attractive to specify the spectral density of idm(t) than to specify the spectral density of the uncorrelated noise current sources, because it is sqrt(2) times smaller.
This means that Linear Technology is absolutely correct in stating that 1 pA/sqrt(Hz) with two 10 kohm source resistors gives you 20 nV/sqrt(Hz), because that 1 pA/sqrt(Hz) is the spectral density of idm(t) and idm(t) is by definition equal but opposite in the inputs.
CONCLUSION
It is all very similar to the NS, the Dutch railways. They are also very good at giving their customers useless information that sounds better than the information they are looking for, and they also assume that all their customers are morons.
PHYSICAL
If you look at what physically happens inside the op-amp, you will see that there are many random processes going on that contribute to the noise current. Some of these, such as the base current shot noise of the input transistors and the collector current shot noise of the output transistors of the base current compensation circuit, are independent and hence uncorrelated for the positive and the negative input. Other contributions, such as those from most other noise processes inside the base current compensation circuit, end up essentially the same in both inputs and cancel when the inputs are driven from equal impedances. (These latter contributions are quite large because the model transistor in the base current compensation circuit runs at a much lower current than the real input transistors, and an amplifying PNP mirror has to amplify the base current up to the correct level.) So far, so good.
MATHEMATICAL
If you look at it from a higher level of abstraction, you see a thing with a current i1(t) flowing into the positive input and a current i2(t) flowing into the negative input. This can be mathematically represented as two current sources with value icm(t) = (i1(t) + i2(t))/2, one from each input to ground (or supply), plus a differential current idm(t) = (i1(t) - i2(t))/2 flowing from the positive to the negative pin. This is just a matter of adding and subtracting, it is not a physical model of what's happening inside the op-amp.
When you want to relate this to the physical representation, you will find that the noise sources that inject independent noise currents into the positive and negative inputs end up in icm(t) as well as in idm(t). In fact 1/sqrt(2) of their spectral density (half their power spectral density) ends up in the spectral density of icm and 1/sqrt(2) in the spectral density of idm.
COMMERCIAL
The honest thing to do for Linear Technology would be to specify the noise current with very different driving impedances for the positive and negative inputs, because most real-life low-noise op-amp circuits have very different driving impedances for the positive and negative inputs. The fact that they don't put 3 pA/sqrt(Hz) or 3.25 pA/sqrt(Hz) or whatever it may be in the tables shows that they've been looking for tricks to get smaller noise current numbers into their datasheet.
If the aim is to minimize the number without risk of getting sued, it is more attractive to specify the spectral density of idm(t) than to specify the spectral density of the uncorrelated noise current sources, because it is sqrt(2) times smaller.
This means that Linear Technology is absolutely correct in stating that 1 pA/sqrt(Hz) with two 10 kohm source resistors gives you 20 nV/sqrt(Hz), because that 1 pA/sqrt(Hz) is the spectral density of idm(t) and idm(t) is by definition equal but opposite in the inputs.
CONCLUSION
It is all very similar to the NS, the Dutch railways. They are also very good at giving their customers useless information that sounds better than the information they are looking for, and they also assume that all their customers are morons.
Last edited:
Just a bit....
However, luckily this OT program begun only after some good ideas regarding the thread topic have been shared.
I suggest we put the current noise lid back on the compensation pot.
Hans
Good idea.
This whole discussion started with plain RIAA-corrected amplifiers, then amplifiers that have RIAA and cartridge response correction, then amplifiers that have RIAA and cartridge response correction and deliberately damp the electrical resonance more than usual, and then we ended up in the off-topic part about noise optimization.
Getting back to measuring phono stage accuracy, does anyone know a way to measure the response of the cartridge and amplifier together other than simply playing a test record and hoping it was recorded accurately?
I don't see a definitive way without including the vinyl/stylus interface. B&K used a precision accelerometer in reverse as a stimulus, but I don't remember the details.
Never got around to trying it, but I'd think any small piezoelectric disc would be a good exciter for a cartridge, especially if you glued a small piece of a record to it, so the stylus was in a groove. A small disk can have a self resonant frequency in the hundreds of kHz or higher.
When I tried them the response was not too flat. There are precision piezo-positioners (some <$40) with self resonances as high as 250k. Piezoelectric Actuators - Thorlabs
Looking over the specs they might be impossible to use at 20kHz since the capacitances are huge. The app notes might ave some good ideas for linearizing cheap disks.
Last edited:
MarcelvdG, I am going to try a cheap in ear headphone speaker glued to a piece of aluminum then place the needle on it with the turntable unplugged. measure the results driving the speaker from an iphone signal generator.
I will see if it requires an RIAA eq. network between the iphone and 32 ohm speaker.
if the cone holds .5 to 1.5 grams it should work
I will see if it requires an RIAA eq. network between the iphone and 32 ohm speaker.
if the cone holds .5 to 1.5 grams it should work
Looking at the smallest 2AB part with 9nF, ie Z=884, this shouldn't be too hard.
The displacement has hysteresis, which means there are some large distortions to be accounted for
Looking at the MM cart resonance plots, I see a relatively small peak compared the the rest of the cart response, after which it drops off very quickly. I would have thought that this would actually be a good thing wrt clicks and pops which have a lot of HF energy whereas there is very little music energy in the top octave. I wonder if the anecdotal evidence from some quarters that certain carts seem quieter wrt clicks and pops has something to do with the response above 15 kHz, where are a bit of peaking, it drops off dramatically. The RIAA curve after 2120 Hz drops off at c 20 dB/ decade anyway.
I find it astounding that an LP sounds as good as it does given the mediums response anomalies. I have some fabulous Ella Fitzgerald recordings - they seem better to my ear than the CD recordings.
I find it astounding that an LP sounds as good as it does given the mediums response anomalies. I have some fabulous Ella Fitzgerald recordings - they seem better to my ear than the CD recordings.
The peak is sometimes small and sometimes rather large. But the peak often comes lower that one would want along with some significant audioband phase shifts.
If you look here you can see that a misloaded (but likely load in many systems) 2M black is 2dB up at 7kHz. New Lamps for Old. That would seems to be a highly audible thing?
If you look here you can see that a misloaded (but likely load in many systems) 2M black is 2dB up at 7kHz. New Lamps for Old. That would seems to be a highly audible thing?
Clearly you would want it as flat as possible. The EQ side in the phono amp is easy to get to 0.2 dB 20-20. The cart real world response is only one part of the equation I think. What was the EQ like in the studio (mics, engineers preferences etc)? CD has the same set of issues in this part of the chain.
The Rega Excel cart in the Jim Lesurf article was bad - Shure and Ortofon seem quite good with the standard loading.
The Rega Excel cart in the Jim Lesurf article was bad - Shure and Ortofon seem quite good with the standard loading.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.