The proof of the pudding is in the eating.
The image below shows A-Weighting left and ITU-R weighting to the right, all 1/3 octave spectra, both for cooled and not cooled.
Up to 2Khz the cooled version is worse with both weighting methods.
Above 2 Khz, cooled has less noise.
Interpretation to what is to be preferred is up to the reader.
Hans
Attachments
Back to the Vinyltrak that Bill mentioned.
In the image below, I have simulated a version with the normally used 47k//200pF Cart load, a version with a 25K Cart load giving a pole at 8 Khz as Bob suggested with his Vinyltrak and a third version with a Cart load of 6K//200pF, giving a pole at 75usec.
All versions used the same circuit based on the single OPA1652, having 3.8 nV/rtHz.
I used a 500mH and 600 Ohm Cart.
In the upper right corner you see the FR for all three, as a prove that they all have the same gain. As expected, the lower the Cart load resistor, the better the FR.
The other three pictures are showing the A-Weighted noise for all three versions from 20Hz up to 20 Khz.
Resp 534nV, 666nV and 1298 nV, meaning that the version with the 8K pole has a noise penalty of 1.9 dB S/N and the version with a pole at 75usec even a noise penalty of 7.7 dB S/N.
That of course is the reason why Bob stopped at 8 Khz where a larger bandwidth was traded for a relatively still acceptable loss in S/N.
S/N referred to 5mV@1Khz for the three versions with cart connected is resp 79.4 dBA, 77,5 dBA and 71,7 dBA.
Just to compare S/N of this simple design with factory made equipment that is always specified with shorted input : 84.0 dBA.
There aren't that many amps on the market with better S/N figures.
As a next step I will compare these results to the Aurak design presenting almost a zero input impedance to the Cart.
Hans
In the image below, I have simulated a version with the normally used 47k//200pF Cart load, a version with a 25K Cart load giving a pole at 8 Khz as Bob suggested with his Vinyltrak and a third version with a Cart load of 6K//200pF, giving a pole at 75usec.
All versions used the same circuit based on the single OPA1652, having 3.8 nV/rtHz.
I used a 500mH and 600 Ohm Cart.
In the upper right corner you see the FR for all three, as a prove that they all have the same gain. As expected, the lower the Cart load resistor, the better the FR.
The other three pictures are showing the A-Weighted noise for all three versions from 20Hz up to 20 Khz.
Resp 534nV, 666nV and 1298 nV, meaning that the version with the 8K pole has a noise penalty of 1.9 dB S/N and the version with a pole at 75usec even a noise penalty of 7.7 dB S/N.
That of course is the reason why Bob stopped at 8 Khz where a larger bandwidth was traded for a relatively still acceptable loss in S/N.
S/N referred to 5mV@1Khz for the three versions with cart connected is resp 79.4 dBA, 77,5 dBA and 71,7 dBA.
Just to compare S/N of this simple design with factory made equipment that is always specified with shorted input : 84.0 dBA.
There aren't that many amps on the market with better S/N figures.
As a next step I will compare these results to the Aurak design presenting almost a zero input impedance to the Cart.
Hans
Attachments
Those are close to Bob's numbers from here http://www.cordellaudio.com/preamplifiers/vinyltrak.shtml . Will be interested in your numbers for the coolect 75uS pole.
Not sure whether I understand your posting correct, but the image below shows the input signal attenuation when switching from 47K to 25K in order to get a 8Khz pole.
Attachments
Last edited:
I simulated 3 versions of a Riaa preamp, again with 500mH + 600 Ohm Cart and OPA1652.
The 3 versions differed in their Cart load resistances, resp 47K//200pF, 25K//200pF for a 8Khz pole and 6k//200pF for a 75usec pole.
All 3 versions were simulated with and without a "cooled" resistor, in this case by using a second OPA1652 with a gain of 10x.
This time I corrected the switched inputs of posting #152, giving slightly different results, but not much.
I won't display all images, that would a bit too much, but the final results are here with minus figures in red:
-------------Noise penalty-------"Cool" Gain---------Overall Gain/Loss
Version 47k ----- 0dBA ----------- 1.6dBA --------------- 1.6dBA
Version 25K --- 2.0dBA ----------- 3.1dBA --------------- 1.1dBA
Version 6K -----7.6dBA ----------- 7.3dBA --------------- 0.3dBA
The big surprise here is how much version-3 with the 75usec pole improves after having applied this "cooled" Cart load resistor.
It is the first time I see that this addition can make a real difference for a Riaa preamp.
Because I used a 500mH+600 Ohm Cart, figures can be slightly different for other Carts.
And also the used One-Amp topology with a OPA1652 and a 330 Ohm feedback resistor has it's own signature.
Hans
The 3 versions differed in their Cart load resistances, resp 47K//200pF, 25K//200pF for a 8Khz pole and 6k//200pF for a 75usec pole.
All 3 versions were simulated with and without a "cooled" resistor, in this case by using a second OPA1652 with a gain of 10x.
This time I corrected the switched inputs of posting #152, giving slightly different results, but not much.
I won't display all images, that would a bit too much, but the final results are here with minus figures in red:
-------------Noise penalty-------"Cool" Gain---------Overall Gain/Loss
Version 47k ----- 0dBA ----------- 1.6dBA --------------- 1.6dBA
Version 25K --- 2.0dBA ----------- 3.1dBA --------------- 1.1dBA
Version 6K -----7.6dBA ----------- 7.3dBA --------------- 0.3dBA
The big surprise here is how much version-3 with the 75usec pole improves after having applied this "cooled" Cart load resistor.
It is the first time I see that this addition can make a real difference for a Riaa preamp.
Because I used a 500mH+600 Ohm Cart, figures can be slightly different for other Carts.
And also the used One-Amp topology with a OPA1652 and a 330 Ohm feedback resistor has it's own signature.
Hans
Marcel, I will come back to your question, but I'm running a bit out of time at the moment.
Anyhow, fired by the succes of version-3 in the posting #169 above, I have made a few extra modifications resulting in a version-3a.
1) I increased the feedback amp's gain from 10x to 15x.
2) I replaced the feedback amp by a LT1028, a low noise bipolar amp with 0.85nV/rtHz and 1pA/rtHz.
With these two mod's, noise went further down by 1.6dB.
Results for this modded version are now:
Version 6K-mod -----7.6dBA ----------- 8.9dBA --------------- 1.3dBA
To conclude:
Implementing the 75usec pole by means of a Cart load resistor and using a "cooled" resistor topology, brings the following benefits over a straightforward design with a 47K load.
1) Largely extended FR
2) Hardly dependant on the TT's cable capacitance.
3) Giving a possible S/N improvement over the standard, in this very example of 1.3 dB.
The only drawback is the increased number of components.
Hans
Anyhow, fired by the succes of version-3 in the posting #169 above, I have made a few extra modifications resulting in a version-3a.
1) I increased the feedback amp's gain from 10x to 15x.
2) I replaced the feedback amp by a LT1028, a low noise bipolar amp with 0.85nV/rtHz and 1pA/rtHz.
With these two mod's, noise went further down by 1.6dB.
Results for this modded version are now:
Version 6K-mod -----7.6dBA ----------- 8.9dBA --------------- 1.3dBA
To conclude:
Implementing the 75usec pole by means of a Cart load resistor and using a "cooled" resistor topology, brings the following benefits over a straightforward design with a 47K load.
1) Largely extended FR
2) Hardly dependant on the TT's cable capacitance.
3) Giving a possible S/N improvement over the standard, in this very example of 1.3 dB.
The only drawback is the increased number of components.
Hans
Attachments
Last edited:
It doesn't matter much in this case, but just for the record: when you read the small print of the LT1028 datasheet, you find that its noise current is 3.25 pA/sqrt(Hz). That 1 pA/sqrt(Hz) only applies with equal impedances driving the positive and negative inputs. The cause is a base current compensation circuit that injects large correlated noise currents into the inputs.
I still had to compare the Aurak to the circuit in posting #171, the optimised version with a Cart load resistor that replaced the usual 75usec in the Riaa network.
Fr of this circuit was shown in posting #165.
The Aurak design goes even a step further by using the zero impedance input of an OPA and putting a pole in this case at 318Hz.
The input OPA corrects this to the point where, in combination with the Cart load, a 75usec pole is created.
Below is shown a full blown optimised differential Aurak, even including servo control to reduce DC offset.
For a current input design, a differential input does not reduce S/N, different to the 3dB S/N penalty when using a differential voltage input design.
The same 500mH + 600 Ohm Cart is used as in the "cooled" resistor version in posting #171 as is the same OPA1652.
S/N for this differential Aurak version is 79.8dBA versus 81.0dBA for the "cooled" version, only 1.2dBA apart, both ridiculously good.
As can be seen, FR is straight up to 100Khz, with less than 3 degrees phase shift at 20 Khz.
All the above simulations have only touched the electrical resonance by the Cart's coil and not the mechanical part.
But so far, for me the differential Aurak is to be preferred above the others.
Hans
Fr of this circuit was shown in posting #165.
The Aurak design goes even a step further by using the zero impedance input of an OPA and putting a pole in this case at 318Hz.
The input OPA corrects this to the point where, in combination with the Cart load, a 75usec pole is created.
Below is shown a full blown optimised differential Aurak, even including servo control to reduce DC offset.
For a current input design, a differential input does not reduce S/N, different to the 3dB S/N penalty when using a differential voltage input design.
The same 500mH + 600 Ohm Cart is used as in the "cooled" resistor version in posting #171 as is the same OPA1652.
S/N for this differential Aurak version is 79.8dBA versus 81.0dBA for the "cooled" version, only 1.2dBA apart, both ridiculously good.
As can be seen, FR is straight up to 100Khz, with less than 3 degrees phase shift at 20 Khz.
All the above simulations have only touched the electrical resonance by the Cart's coil and not the mechanical part.
But so far, for me the differential Aurak is to be preferred above the others.
Hans
Attachments
Last edited:
Marcel, here are the figures that you asked for.
A-Weighted
Non Cooled 531nV
Cooled 437nV
ITU-R
Non Cooled 1420nV
Cooled 896nV
Hans
A few remarks to my posting #173.
1) When I click on the image to enlarge it, I get a coloured band through the image on my Windows system but not on my iPad.
If needed I can republish it.
2) C3 should be 4p7 and not 4n7 as in the image
3) The servo is not just to reduce DC offset, but more important to suppress Common Mode.
Hans
1) When I click on the image to enlarge it, I get a coloured band through the image on my Windows system but not on my iPad.
If needed I can republish it.
2) C3 should be 4p7 and not 4n7 as in the image
3) The servo is not just to reduce DC offset, but more important to suppress Common Mode.
Hans
OK, so 1.69 dB improvement with A-weighting and 4 dB improvement with ITU-R 468 weighting (assuming that the ITU-R quasi-peak meter would give the same ratio as an RMS measurement). Great, that qualitatively matches my calculations, so I won't need to go looking for mistakes.
Using my rules of thumb and taking into account only 3.8 nV/sqrt(Hz) per op-amp and the thermal noise of the 600 ohm cartridge resistance, 330 ohm feedback resistor and 47 kohm or 11*47 kohm termination resistor, I find these values:
A-weighted:
non-cooled 497.4 nV RMS
cooled 395.1 nV RMS
improvement 2 dB
ITU-R 468
non-cooled 1.375 uV RMS
cooled 899.3 nV RMS
improvement 3.69 dB
My values are a bit off because I neglect some things, but they definitely show the right trends.
A-weighted:
non-cooled 497.4 nV RMS
cooled 395.1 nV RMS
improvement 2 dB
ITU-R 468
non-cooled 1.375 uV RMS
cooled 899.3 nV RMS
improvement 3.69 dB
My values are a bit off because I neglect some things, but they definitely show the right trends.
Marcel,
Coming back to this issue, it is true that input current compensation makes that for matched versus non matched input resistances, different current noise values should be used, because correlated noise will be cancelled by CMRR in matched Rs versions.
However, looking at the images that Linear supplies, I find it impossible to match their shapes.
There are two relevant points in these graphs, the beginning at 1 Ohm where voltage noise is dominant and at 10Kohm where current noise is dominant.
Both matched and unmatched versions start at 0.85nV/rtHz for 1Khz noise and both end at 33nV/rtHz for noise at 1Khz.
Matched resistors.
When using both reference points at start and end, this can only be done when using a value of 2.0pA/rtHz and not 1.0pA/rtHz.
This "matched" graph looks quite identical to Linear's graph.
Apart from the double current noise figure, so far so good.
Unmatched resistors
When swiching to the unmatched version, an extra uncorrelated current has to be taken into account. I have called this current i2.
Matching the end point at 33nV/rtHz is now possible for i2=2.2pA/rtHz.
The rest of the graph however differs from Linear's version.
So I don't understand what Linear did to produce these files, but I have a problem with them, matched noise at 2.0 instead of 1.0 pA/rtHz and a graph for the unmatched version that is not identical.
However, your statement with 1pA/rtHz matched and 3.25pA/rtHz unmatched seems not to be confirmed by these graph's.
Hans
Attachments
I only looked at the end point of the unmatched curve, which should be the most sensitive to the op-amp's noise current, and read it off a bit more pessimistically than you:
35 nV/sqrt(Hz) -> 1225E-18 V^2/Hz
Op-amp's voltage noise: 0.85 nV/sqrt(Hz) -> 0.7225E-18 V^2/Hz (from the tables in the datasheet)
Thermal noise of 10 kohm at 25 degrees C: 164.656319E-18 V^2/Hz
Difference: 1059.621181E-18 V^2/Hz, or 32.55182301 nV/sqrt(Hz)
Divided by 10 kohm: 3.255182301 pA/sqrt(Hz)
If the end point is 33 nV/sqrt(Hz), it becomes 3.039113655 pA/sqrt(Hz).
This current is the total noise current you need to take into account for very unmatched impedances, so it includes the correlated and the uncorrelated noise currents. I never tried to split it up, or to match the rest of the curves. The 1 pA/sqrt(Hz) I mentioned for matched impedances comes straight from the tables in the datasheet, not from the graphs.
35 nV/sqrt(Hz) -> 1225E-18 V^2/Hz
Op-amp's voltage noise: 0.85 nV/sqrt(Hz) -> 0.7225E-18 V^2/Hz (from the tables in the datasheet)
Thermal noise of 10 kohm at 25 degrees C: 164.656319E-18 V^2/Hz
Difference: 1059.621181E-18 V^2/Hz, or 32.55182301 nV/sqrt(Hz)
Divided by 10 kohm: 3.255182301 pA/sqrt(Hz)
If the end point is 33 nV/sqrt(Hz), it becomes 3.039113655 pA/sqrt(Hz).
This current is the total noise current you need to take into account for very unmatched impedances, so it includes the correlated and the uncorrelated noise currents. I never tried to split it up, or to match the rest of the curves. The 1 pA/sqrt(Hz) I mentioned for matched impedances comes straight from the tables in the datasheet, not from the graphs.
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.
Make the calculation for the matched Rs graph at the left and you will find a current noise of 2.0pA/rtHz.
So either the graphs are incorrect or the 1.0pA/rtHz in the specs is incorrect.
For the time being I suspect the graphs.
Hans
) when using 33nV/rtHz@1Khz.Make the calculation for the matched Rs graph at the left and you will find a current noise of 2.0pA/rtHz.
So either the graphs are incorrect or the 1.0pA/rtHz in the specs is incorrect.
For the time being I suspect the graphs.
Hans
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.